program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3402.3.2"}, {"coremlc-version", "3402.5.1"}})]
{
    func main<ios17>(tensor<fp16, [1, 80, 3000]> logmel_data) {
            tensor<int32, [3]> input_1_begin_0 = const()[name = tensor<string, []>("input_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> input_1_end_0 = const()[name = tensor<string, []>("input_1_end_0"), val = tensor<int32, [3]>([1, 80, 1200])];
            tensor<bool, [3]> input_1_end_mask_0 = const()[name = tensor<string, []>("input_1_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [1, 80, 1200]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = logmel_data)[name = tensor<string, []>("input_1_cast_fp16")];
            tensor<string, []> var_84_pad_type_0 = const()[name = tensor<string, []>("op_84_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_84_pad_0 = const()[name = tensor<string, []>("op_84_pad_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [1]> var_84_strides_0 = const()[name = tensor<string, []>("op_84_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_84_dilations_0 = const()[name = tensor<string, []>("op_84_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> var_84_groups_0 = const()[name = tensor<string, []>("op_84_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 80, 3]> const_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [122880]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123008))), name = tensor<string, []>("const_0_to_fp16_palettized"), shape = tensor<uint32, [3]>([1024, 80, 3])];
            tensor<fp16, [1024]> const_1_to_fp16 = const()[name = tensor<string, []>("const_1_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123136)))];
            tensor<fp16, [1, 1024, 1200]> var_84_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_84_dilations_0, groups = var_84_groups_0, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_84_strides_0, weight = const_0_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor<string, []>("op_84_cast_fp16")];
            tensor<string, []> input_3_mode_0 = const()[name = tensor<string, []>("input_3_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 1200]> input_3_cast_fp16 = gelu(mode = input_3_mode_0, x = var_84_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<string, []> var_106_pad_type_0 = const()[name = tensor<string, []>("op_106_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_106_pad_0 = const()[name = tensor<string, []>("op_106_pad_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [1]> var_106_strides_0 = const()[name = tensor<string, []>("op_106_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> var_106_dilations_0 = const()[name = tensor<string, []>("op_106_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> var_106_groups_0 = const()[name = tensor<string, []>("op_106_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 3]> const_2_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1572864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125248))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1698176))), name = tensor<string, []>("const_2_to_fp16_palettized"), shape = tensor<uint32, [3]>([1024, 1024, 3])];
            tensor<fp16, [1024]> const_3_to_fp16 = const()[name = tensor<string, []>("const_3_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1698304)))];
            tensor<fp16, [1, 1024, 600]> var_106_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_106_dilations_0, groups = var_106_groups_0, pad = var_106_pad_0, pad_type = var_106_pad_type_0, strides = var_106_strides_0, weight = const_2_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor<string, []>("op_106_cast_fp16")];
            tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 600]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_106_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
            tensor<fp16, [1, 1024, 600]> pos_emb_1_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1700416))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2007680))), name = tensor<string, []>("pos_emb_1_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 1024, 600])];
            tensor<fp16, [1, 1024, 600]> var_132_cast_fp16 = add(x = x_3_cast_fp16, y = pos_emb_1_to_fp16_palettized)[name = tensor<string, []>("op_132_cast_fp16")];
            tensor<int32, [1]> inputs_1_axes_0 = const()[name = tensor<string, []>("inputs_1_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1024, 1, 600]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_132_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")];
            tensor<int32, []> var_147 = const()[name = tensor<string, []>("op_147"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_5_axes_0 = const()[name = tensor<string, []>("input_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_5_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2007808)))];
            tensor<fp16, [1024]> input_5_beta_0_to_fp16 = const()[name = tensor<string, []>("input_5_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2009920)))];
            tensor<fp16, []> var_163_to_fp16 = const()[name = tensor<string, []>("op_163_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_5_cast_fp16 = layer_norm(axes = input_5_axes_0, beta = input_5_beta_0_to_fp16, epsilon = var_163_to_fp16, gamma = input_5_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<string, []> q_1_pad_type_0 = const()[name = tensor<string, []>("q_1_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_1_strides_0 = const()[name = tensor<string, []>("q_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_1_pad_0 = const()[name = tensor<string, []>("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_1_dilations_0 = const()[name = tensor<string, []>("q_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_1_groups_0 = const()[name = tensor<string, []>("q_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_198_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2012032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2536384))), name = tensor<string, []>("op_198_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_198_bias_0_to_fp16 = const()[name = tensor<string, []>("op_198_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2536512)))];
            tensor<fp16, [1, 1024, 1, 600]> var_198_cast_fp16 = conv(bias = var_198_bias_0_to_fp16, dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = op_198_weight_0_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor<string, []>("op_198_cast_fp16")];
            tensor<string, []> k_1_pad_type_0 = const()[name = tensor<string, []>("k_1_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_1_strides_0 = const()[name = tensor<string, []>("k_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_1_pad_0 = const()[name = tensor<string, []>("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_1_dilations_0 = const()[name = tensor<string, []>("k_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_1_groups_0 = const()[name = tensor<string, []>("k_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_0_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2538624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3062976))), name = tensor<string, []>("blocks_0_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = blocks_0_attn_key_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor<string, []>("k_1_cast_fp16")];
            tensor<string, []> var_196_pad_type_0 = const()[name = tensor<string, []>("op_196_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_196_strides_0 = const()[name = tensor<string, []>("op_196_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_196_pad_0 = const()[name = tensor<string, []>("op_196_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_196_dilations_0 = const()[name = tensor<string, []>("op_196_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_196_groups_0 = const()[name = tensor<string, []>("op_196_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_0_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3063104))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3587456))), name = tensor<string, []>("blocks_0_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_0_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3587584)))];
            tensor<fp16, [1, 1024, 1, 600]> var_196_cast_fp16 = conv(bias = blocks_0_attn_value_bias_to_fp16, dilations = var_196_dilations_0, groups = var_196_groups_0, pad = var_196_pad_0, pad_type = var_196_pad_type_0, strides = var_196_strides_0, weight = blocks_0_attn_value_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor<string, []>("op_196_cast_fp16")];
            tensor<int32, [16]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_199_axis_0 = const()[name = tensor<string, []>("op_199_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_199_cast_fp16_15 = split(axis = var_199_axis_0, split_sizes = tile_0, x = var_198_cast_fp16)[name = tensor<string, []>("op_199_cast_fp16")];
            tensor<int32, [4]> var_216_perm_0 = const()[name = tensor<string, []>("op_216_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_217_axis_0 = const()[name = tensor<string, []>("op_217_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_216_cast_fp16 = transpose(perm = var_216_perm_0, x = k_1_cast_fp16)[name = tensor<string, []>("transpose_35")];
            tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_217_cast_fp16_15 = split(axis = var_217_axis_0, split_sizes = tile_1, x = var_216_cast_fp16)[name = tensor<string, []>("op_217_cast_fp16")];
            tensor<int32, [16]> tile_2 = const()[name = tensor<string, []>("tile_2"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_234_axis_0 = const()[name = tensor<string, []>("op_234_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_234_cast_fp16_15 = split(axis = var_234_axis_0, split_sizes = tile_2, x = var_196_cast_fp16)[name = tensor<string, []>("op_234_cast_fp16")];
            tensor<string, []> aw_1_equation_0 = const()[name = tensor<string, []>("aw_1_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_1_cast_fp16 = einsum(equation = aw_1_equation_0, values = (var_217_cast_fp16_0, var_199_cast_fp16_0))[name = tensor<string, []>("aw_1_cast_fp16")];
            tensor<string, []> aw_3_equation_0 = const()[name = tensor<string, []>("aw_3_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_3_cast_fp16 = einsum(equation = aw_3_equation_0, values = (var_217_cast_fp16_1, var_199_cast_fp16_1))[name = tensor<string, []>("aw_3_cast_fp16")];
            tensor<string, []> aw_5_equation_0 = const()[name = tensor<string, []>("aw_5_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_5_cast_fp16 = einsum(equation = aw_5_equation_0, values = (var_217_cast_fp16_2, var_199_cast_fp16_2))[name = tensor<string, []>("aw_5_cast_fp16")];
            tensor<string, []> aw_7_equation_0 = const()[name = tensor<string, []>("aw_7_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_7_cast_fp16 = einsum(equation = aw_7_equation_0, values = (var_217_cast_fp16_3, var_199_cast_fp16_3))[name = tensor<string, []>("aw_7_cast_fp16")];
            tensor<string, []> aw_9_equation_0 = const()[name = tensor<string, []>("aw_9_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_9_cast_fp16 = einsum(equation = aw_9_equation_0, values = (var_217_cast_fp16_4, var_199_cast_fp16_4))[name = tensor<string, []>("aw_9_cast_fp16")];
            tensor<string, []> aw_11_equation_0 = const()[name = tensor<string, []>("aw_11_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_11_cast_fp16 = einsum(equation = aw_11_equation_0, values = (var_217_cast_fp16_5, var_199_cast_fp16_5))[name = tensor<string, []>("aw_11_cast_fp16")];
            tensor<string, []> aw_13_equation_0 = const()[name = tensor<string, []>("aw_13_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_13_cast_fp16 = einsum(equation = aw_13_equation_0, values = (var_217_cast_fp16_6, var_199_cast_fp16_6))[name = tensor<string, []>("aw_13_cast_fp16")];
            tensor<string, []> aw_15_equation_0 = const()[name = tensor<string, []>("aw_15_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_15_cast_fp16 = einsum(equation = aw_15_equation_0, values = (var_217_cast_fp16_7, var_199_cast_fp16_7))[name = tensor<string, []>("aw_15_cast_fp16")];
            tensor<string, []> aw_17_equation_0 = const()[name = tensor<string, []>("aw_17_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_17_cast_fp16 = einsum(equation = aw_17_equation_0, values = (var_217_cast_fp16_8, var_199_cast_fp16_8))[name = tensor<string, []>("aw_17_cast_fp16")];
            tensor<string, []> aw_19_equation_0 = const()[name = tensor<string, []>("aw_19_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_19_cast_fp16 = einsum(equation = aw_19_equation_0, values = (var_217_cast_fp16_9, var_199_cast_fp16_9))[name = tensor<string, []>("aw_19_cast_fp16")];
            tensor<string, []> aw_21_equation_0 = const()[name = tensor<string, []>("aw_21_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_21_cast_fp16 = einsum(equation = aw_21_equation_0, values = (var_217_cast_fp16_10, var_199_cast_fp16_10))[name = tensor<string, []>("aw_21_cast_fp16")];
            tensor<string, []> aw_23_equation_0 = const()[name = tensor<string, []>("aw_23_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_23_cast_fp16 = einsum(equation = aw_23_equation_0, values = (var_217_cast_fp16_11, var_199_cast_fp16_11))[name = tensor<string, []>("aw_23_cast_fp16")];
            tensor<string, []> aw_25_equation_0 = const()[name = tensor<string, []>("aw_25_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_25_cast_fp16 = einsum(equation = aw_25_equation_0, values = (var_217_cast_fp16_12, var_199_cast_fp16_12))[name = tensor<string, []>("aw_25_cast_fp16")];
            tensor<string, []> aw_27_equation_0 = const()[name = tensor<string, []>("aw_27_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_27_cast_fp16 = einsum(equation = aw_27_equation_0, values = (var_217_cast_fp16_13, var_199_cast_fp16_13))[name = tensor<string, []>("aw_27_cast_fp16")];
            tensor<string, []> aw_29_equation_0 = const()[name = tensor<string, []>("aw_29_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_29_cast_fp16 = einsum(equation = aw_29_equation_0, values = (var_217_cast_fp16_14, var_199_cast_fp16_14))[name = tensor<string, []>("aw_29_cast_fp16")];
            tensor<string, []> aw_31_equation_0 = const()[name = tensor<string, []>("aw_31_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_31_cast_fp16 = einsum(equation = aw_31_equation_0, values = (var_217_cast_fp16_15, var_199_cast_fp16_15))[name = tensor<string, []>("aw_31_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_283_cast_fp16 = softmax(axis = var_147, x = aw_1_cast_fp16)[name = tensor<string, []>("op_283_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_284_cast_fp16 = softmax(axis = var_147, x = aw_3_cast_fp16)[name = tensor<string, []>("op_284_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_285_cast_fp16 = softmax(axis = var_147, x = aw_5_cast_fp16)[name = tensor<string, []>("op_285_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_286_cast_fp16 = softmax(axis = var_147, x = aw_7_cast_fp16)[name = tensor<string, []>("op_286_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_287_cast_fp16 = softmax(axis = var_147, x = aw_9_cast_fp16)[name = tensor<string, []>("op_287_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_288_cast_fp16 = softmax(axis = var_147, x = aw_11_cast_fp16)[name = tensor<string, []>("op_288_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_289_cast_fp16 = softmax(axis = var_147, x = aw_13_cast_fp16)[name = tensor<string, []>("op_289_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_290_cast_fp16 = softmax(axis = var_147, x = aw_15_cast_fp16)[name = tensor<string, []>("op_290_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_291_cast_fp16 = softmax(axis = var_147, x = aw_17_cast_fp16)[name = tensor<string, []>("op_291_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_292_cast_fp16 = softmax(axis = var_147, x = aw_19_cast_fp16)[name = tensor<string, []>("op_292_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_293_cast_fp16 = softmax(axis = var_147, x = aw_21_cast_fp16)[name = tensor<string, []>("op_293_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_294_cast_fp16 = softmax(axis = var_147, x = aw_23_cast_fp16)[name = tensor<string, []>("op_294_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_295_cast_fp16 = softmax(axis = var_147, x = aw_25_cast_fp16)[name = tensor<string, []>("op_295_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_296_cast_fp16 = softmax(axis = var_147, x = aw_27_cast_fp16)[name = tensor<string, []>("op_296_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_297_cast_fp16 = softmax(axis = var_147, x = aw_29_cast_fp16)[name = tensor<string, []>("op_297_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_298_cast_fp16 = softmax(axis = var_147, x = aw_31_cast_fp16)[name = tensor<string, []>("op_298_cast_fp16")];
            tensor<string, []> var_300_equation_0 = const()[name = tensor<string, []>("op_300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_300_cast_fp16 = einsum(equation = var_300_equation_0, values = (var_234_cast_fp16_0, var_283_cast_fp16))[name = tensor<string, []>("op_300_cast_fp16")];
            tensor<string, []> var_302_equation_0 = const()[name = tensor<string, []>("op_302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_302_cast_fp16 = einsum(equation = var_302_equation_0, values = (var_234_cast_fp16_1, var_284_cast_fp16))[name = tensor<string, []>("op_302_cast_fp16")];
            tensor<string, []> var_304_equation_0 = const()[name = tensor<string, []>("op_304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_304_cast_fp16 = einsum(equation = var_304_equation_0, values = (var_234_cast_fp16_2, var_285_cast_fp16))[name = tensor<string, []>("op_304_cast_fp16")];
            tensor<string, []> var_306_equation_0 = const()[name = tensor<string, []>("op_306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_306_cast_fp16 = einsum(equation = var_306_equation_0, values = (var_234_cast_fp16_3, var_286_cast_fp16))[name = tensor<string, []>("op_306_cast_fp16")];
            tensor<string, []> var_308_equation_0 = const()[name = tensor<string, []>("op_308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_308_cast_fp16 = einsum(equation = var_308_equation_0, values = (var_234_cast_fp16_4, var_287_cast_fp16))[name = tensor<string, []>("op_308_cast_fp16")];
            tensor<string, []> var_310_equation_0 = const()[name = tensor<string, []>("op_310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_310_cast_fp16 = einsum(equation = var_310_equation_0, values = (var_234_cast_fp16_5, var_288_cast_fp16))[name = tensor<string, []>("op_310_cast_fp16")];
            tensor<string, []> var_312_equation_0 = const()[name = tensor<string, []>("op_312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_312_cast_fp16 = einsum(equation = var_312_equation_0, values = (var_234_cast_fp16_6, var_289_cast_fp16))[name = tensor<string, []>("op_312_cast_fp16")];
            tensor<string, []> var_314_equation_0 = const()[name = tensor<string, []>("op_314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_314_cast_fp16 = einsum(equation = var_314_equation_0, values = (var_234_cast_fp16_7, var_290_cast_fp16))[name = tensor<string, []>("op_314_cast_fp16")];
            tensor<string, []> var_316_equation_0 = const()[name = tensor<string, []>("op_316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_316_cast_fp16 = einsum(equation = var_316_equation_0, values = (var_234_cast_fp16_8, var_291_cast_fp16))[name = tensor<string, []>("op_316_cast_fp16")];
            tensor<string, []> var_318_equation_0 = const()[name = tensor<string, []>("op_318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_318_cast_fp16 = einsum(equation = var_318_equation_0, values = (var_234_cast_fp16_9, var_292_cast_fp16))[name = tensor<string, []>("op_318_cast_fp16")];
            tensor<string, []> var_320_equation_0 = const()[name = tensor<string, []>("op_320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_320_cast_fp16 = einsum(equation = var_320_equation_0, values = (var_234_cast_fp16_10, var_293_cast_fp16))[name = tensor<string, []>("op_320_cast_fp16")];
            tensor<string, []> var_322_equation_0 = const()[name = tensor<string, []>("op_322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_322_cast_fp16 = einsum(equation = var_322_equation_0, values = (var_234_cast_fp16_11, var_294_cast_fp16))[name = tensor<string, []>("op_322_cast_fp16")];
            tensor<string, []> var_324_equation_0 = const()[name = tensor<string, []>("op_324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_324_cast_fp16 = einsum(equation = var_324_equation_0, values = (var_234_cast_fp16_12, var_295_cast_fp16))[name = tensor<string, []>("op_324_cast_fp16")];
            tensor<string, []> var_326_equation_0 = const()[name = tensor<string, []>("op_326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_326_cast_fp16 = einsum(equation = var_326_equation_0, values = (var_234_cast_fp16_13, var_296_cast_fp16))[name = tensor<string, []>("op_326_cast_fp16")];
            tensor<string, []> var_328_equation_0 = const()[name = tensor<string, []>("op_328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_328_cast_fp16 = einsum(equation = var_328_equation_0, values = (var_234_cast_fp16_14, var_297_cast_fp16))[name = tensor<string, []>("op_328_cast_fp16")];
            tensor<string, []> var_330_equation_0 = const()[name = tensor<string, []>("op_330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_330_cast_fp16 = einsum(equation = var_330_equation_0, values = (var_234_cast_fp16_15, var_298_cast_fp16))[name = tensor<string, []>("op_330_cast_fp16")];
            tensor<bool, []> input_7_interleave_0 = const()[name = tensor<string, []>("input_7_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_7_cast_fp16 = concat(axis = var_147, interleave = input_7_interleave_0, values = (var_300_cast_fp16, var_302_cast_fp16, var_304_cast_fp16, var_306_cast_fp16, var_308_cast_fp16, var_310_cast_fp16, var_312_cast_fp16, var_314_cast_fp16, var_316_cast_fp16, var_318_cast_fp16, var_320_cast_fp16, var_322_cast_fp16, var_324_cast_fp16, var_326_cast_fp16, var_328_cast_fp16, var_330_cast_fp16))[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<string, []> var_339_pad_type_0 = const()[name = tensor<string, []>("op_339_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_339_strides_0 = const()[name = tensor<string, []>("op_339_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_339_pad_0 = const()[name = tensor<string, []>("op_339_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_339_dilations_0 = const()[name = tensor<string, []>("op_339_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_339_groups_0 = const()[name = tensor<string, []>("op_339_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_0_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3589696))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4114048))), name = tensor<string, []>("blocks_0_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_0_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4114176)))];
            tensor<fp16, [1, 1024, 1, 600]> var_339_cast_fp16 = conv(bias = blocks_0_attn_out_bias_to_fp16, dilations = var_339_dilations_0, groups = var_339_groups_0, pad = var_339_pad_0, pad_type = var_339_pad_type_0, strides = var_339_strides_0, weight = blocks_0_attn_out_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor<string, []>("op_339_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_339_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")];
            tensor<int32, [1]> input_9_axes_0 = const()[name = tensor<string, []>("input_9_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_9_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4116288)))];
            tensor<fp16, [1024]> input_9_beta_0_to_fp16 = const()[name = tensor<string, []>("input_9_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4118400)))];
            tensor<fp16, []> var_349_to_fp16 = const()[name = tensor<string, []>("op_349_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_9_cast_fp16 = layer_norm(axes = input_9_axes_0, beta = input_9_beta_0_to_fp16, epsilon = var_349_to_fp16, gamma = input_9_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<string, []> input_11_pad_type_0 = const()[name = tensor<string, []>("input_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_11_strides_0 = const()[name = tensor<string, []>("input_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_11_pad_0 = const()[name = tensor<string, []>("input_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_11_dilations_0 = const()[name = tensor<string, []>("input_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_11_groups_0 = const()[name = tensor<string, []>("input_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_0_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4120512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6217728))), name = tensor<string, []>("blocks_0_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_0_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6217856)))];
            tensor<fp16, [1, 4096, 1, 600]> input_11_cast_fp16 = conv(bias = blocks_0_mlp_0_bias_to_fp16, dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = blocks_0_mlp_0_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<string, []> input_13_mode_0 = const()[name = tensor<string, []>("input_13_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_13_cast_fp16 = gelu(mode = input_13_mode_0, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<string, []> var_375_pad_type_0 = const()[name = tensor<string, []>("op_375_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_375_strides_0 = const()[name = tensor<string, []>("op_375_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_375_pad_0 = const()[name = tensor<string, []>("op_375_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_375_dilations_0 = const()[name = tensor<string, []>("op_375_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_375_groups_0 = const()[name = tensor<string, []>("op_375_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_0_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6226112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8323328))), name = tensor<string, []>("blocks_0_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_0_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8323456)))];
            tensor<fp16, [1, 1024, 1, 600]> var_375_cast_fp16 = conv(bias = blocks_0_mlp_2_bias_to_fp16, dilations = var_375_dilations_0, groups = var_375_groups_0, pad = var_375_pad_0, pad_type = var_375_pad_type_0, strides = var_375_strides_0, weight = blocks_0_mlp_2_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("op_375_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = var_375_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")];
            tensor<int32, []> var_384 = const()[name = tensor<string, []>("op_384"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_15_axes_0 = const()[name = tensor<string, []>("input_15_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_15_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8325568)))];
            tensor<fp16, [1024]> input_15_beta_0_to_fp16 = const()[name = tensor<string, []>("input_15_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8327680)))];
            tensor<fp16, []> var_400_to_fp16 = const()[name = tensor<string, []>("op_400_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_15_cast_fp16 = layer_norm(axes = input_15_axes_0, beta = input_15_beta_0_to_fp16, epsilon = var_400_to_fp16, gamma = input_15_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<string, []> q_3_pad_type_0 = const()[name = tensor<string, []>("q_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_3_strides_0 = const()[name = tensor<string, []>("q_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_3_pad_0 = const()[name = tensor<string, []>("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_3_dilations_0 = const()[name = tensor<string, []>("q_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_3_groups_0 = const()[name = tensor<string, []>("q_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_435_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8329792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8854144))), name = tensor<string, []>("op_435_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_435_bias_0_to_fp16 = const()[name = tensor<string, []>("op_435_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8854272)))];
            tensor<fp16, [1, 1024, 1, 600]> var_435_cast_fp16 = conv(bias = var_435_bias_0_to_fp16, dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = op_435_weight_0_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor<string, []>("op_435_cast_fp16")];
            tensor<string, []> k_3_pad_type_0 = const()[name = tensor<string, []>("k_3_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_3_strides_0 = const()[name = tensor<string, []>("k_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_3_pad_0 = const()[name = tensor<string, []>("k_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_3_dilations_0 = const()[name = tensor<string, []>("k_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_3_groups_0 = const()[name = tensor<string, []>("k_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_1_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8856384))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9380736))), name = tensor<string, []>("blocks_1_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_3_cast_fp16 = conv(dilations = k_3_dilations_0, groups = k_3_groups_0, pad = k_3_pad_0, pad_type = k_3_pad_type_0, strides = k_3_strides_0, weight = blocks_1_attn_key_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor<string, []>("k_3_cast_fp16")];
            tensor<string, []> var_433_pad_type_0 = const()[name = tensor<string, []>("op_433_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_433_strides_0 = const()[name = tensor<string, []>("op_433_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_433_pad_0 = const()[name = tensor<string, []>("op_433_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_433_dilations_0 = const()[name = tensor<string, []>("op_433_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_433_groups_0 = const()[name = tensor<string, []>("op_433_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_1_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9380864))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9905216))), name = tensor<string, []>("blocks_1_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_1_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9905344)))];
            tensor<fp16, [1, 1024, 1, 600]> var_433_cast_fp16 = conv(bias = blocks_1_attn_value_bias_to_fp16, dilations = var_433_dilations_0, groups = var_433_groups_0, pad = var_433_pad_0, pad_type = var_433_pad_type_0, strides = var_433_strides_0, weight = blocks_1_attn_value_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor<string, []>("op_433_cast_fp16")];
            tensor<int32, [16]> tile_3 = const()[name = tensor<string, []>("tile_3"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_436_axis_0 = const()[name = tensor<string, []>("op_436_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_436_cast_fp16_15 = split(axis = var_436_axis_0, split_sizes = tile_3, x = var_435_cast_fp16)[name = tensor<string, []>("op_436_cast_fp16")];
            tensor<int32, [4]> var_453_perm_0 = const()[name = tensor<string, []>("op_453_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_4 = const()[name = tensor<string, []>("tile_4"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_454_axis_0 = const()[name = tensor<string, []>("op_454_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_453_cast_fp16 = transpose(perm = var_453_perm_0, x = k_3_cast_fp16)[name = tensor<string, []>("transpose_34")];
            tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_454_cast_fp16_15 = split(axis = var_454_axis_0, split_sizes = tile_4, x = var_453_cast_fp16)[name = tensor<string, []>("op_454_cast_fp16")];
            tensor<int32, [16]> tile_5 = const()[name = tensor<string, []>("tile_5"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_471_axis_0 = const()[name = tensor<string, []>("op_471_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_471_cast_fp16_15 = split(axis = var_471_axis_0, split_sizes = tile_5, x = var_433_cast_fp16)[name = tensor<string, []>("op_471_cast_fp16")];
            tensor<string, []> aw_33_equation_0 = const()[name = tensor<string, []>("aw_33_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_33_cast_fp16 = einsum(equation = aw_33_equation_0, values = (var_454_cast_fp16_0, var_436_cast_fp16_0))[name = tensor<string, []>("aw_33_cast_fp16")];
            tensor<string, []> aw_35_equation_0 = const()[name = tensor<string, []>("aw_35_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_35_cast_fp16 = einsum(equation = aw_35_equation_0, values = (var_454_cast_fp16_1, var_436_cast_fp16_1))[name = tensor<string, []>("aw_35_cast_fp16")];
            tensor<string, []> aw_37_equation_0 = const()[name = tensor<string, []>("aw_37_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_37_cast_fp16 = einsum(equation = aw_37_equation_0, values = (var_454_cast_fp16_2, var_436_cast_fp16_2))[name = tensor<string, []>("aw_37_cast_fp16")];
            tensor<string, []> aw_39_equation_0 = const()[name = tensor<string, []>("aw_39_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_39_cast_fp16 = einsum(equation = aw_39_equation_0, values = (var_454_cast_fp16_3, var_436_cast_fp16_3))[name = tensor<string, []>("aw_39_cast_fp16")];
            tensor<string, []> aw_41_equation_0 = const()[name = tensor<string, []>("aw_41_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_41_cast_fp16 = einsum(equation = aw_41_equation_0, values = (var_454_cast_fp16_4, var_436_cast_fp16_4))[name = tensor<string, []>("aw_41_cast_fp16")];
            tensor<string, []> aw_43_equation_0 = const()[name = tensor<string, []>("aw_43_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_43_cast_fp16 = einsum(equation = aw_43_equation_0, values = (var_454_cast_fp16_5, var_436_cast_fp16_5))[name = tensor<string, []>("aw_43_cast_fp16")];
            tensor<string, []> aw_45_equation_0 = const()[name = tensor<string, []>("aw_45_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_45_cast_fp16 = einsum(equation = aw_45_equation_0, values = (var_454_cast_fp16_6, var_436_cast_fp16_6))[name = tensor<string, []>("aw_45_cast_fp16")];
            tensor<string, []> aw_47_equation_0 = const()[name = tensor<string, []>("aw_47_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_47_cast_fp16 = einsum(equation = aw_47_equation_0, values = (var_454_cast_fp16_7, var_436_cast_fp16_7))[name = tensor<string, []>("aw_47_cast_fp16")];
            tensor<string, []> aw_49_equation_0 = const()[name = tensor<string, []>("aw_49_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_49_cast_fp16 = einsum(equation = aw_49_equation_0, values = (var_454_cast_fp16_8, var_436_cast_fp16_8))[name = tensor<string, []>("aw_49_cast_fp16")];
            tensor<string, []> aw_51_equation_0 = const()[name = tensor<string, []>("aw_51_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_51_cast_fp16 = einsum(equation = aw_51_equation_0, values = (var_454_cast_fp16_9, var_436_cast_fp16_9))[name = tensor<string, []>("aw_51_cast_fp16")];
            tensor<string, []> aw_53_equation_0 = const()[name = tensor<string, []>("aw_53_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_53_cast_fp16 = einsum(equation = aw_53_equation_0, values = (var_454_cast_fp16_10, var_436_cast_fp16_10))[name = tensor<string, []>("aw_53_cast_fp16")];
            tensor<string, []> aw_55_equation_0 = const()[name = tensor<string, []>("aw_55_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_55_cast_fp16 = einsum(equation = aw_55_equation_0, values = (var_454_cast_fp16_11, var_436_cast_fp16_11))[name = tensor<string, []>("aw_55_cast_fp16")];
            tensor<string, []> aw_57_equation_0 = const()[name = tensor<string, []>("aw_57_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_57_cast_fp16 = einsum(equation = aw_57_equation_0, values = (var_454_cast_fp16_12, var_436_cast_fp16_12))[name = tensor<string, []>("aw_57_cast_fp16")];
            tensor<string, []> aw_59_equation_0 = const()[name = tensor<string, []>("aw_59_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_59_cast_fp16 = einsum(equation = aw_59_equation_0, values = (var_454_cast_fp16_13, var_436_cast_fp16_13))[name = tensor<string, []>("aw_59_cast_fp16")];
            tensor<string, []> aw_61_equation_0 = const()[name = tensor<string, []>("aw_61_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_61_cast_fp16 = einsum(equation = aw_61_equation_0, values = (var_454_cast_fp16_14, var_436_cast_fp16_14))[name = tensor<string, []>("aw_61_cast_fp16")];
            tensor<string, []> aw_63_equation_0 = const()[name = tensor<string, []>("aw_63_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_63_cast_fp16 = einsum(equation = aw_63_equation_0, values = (var_454_cast_fp16_15, var_436_cast_fp16_15))[name = tensor<string, []>("aw_63_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_520_cast_fp16 = softmax(axis = var_384, x = aw_33_cast_fp16)[name = tensor<string, []>("op_520_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_521_cast_fp16 = softmax(axis = var_384, x = aw_35_cast_fp16)[name = tensor<string, []>("op_521_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_522_cast_fp16 = softmax(axis = var_384, x = aw_37_cast_fp16)[name = tensor<string, []>("op_522_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_523_cast_fp16 = softmax(axis = var_384, x = aw_39_cast_fp16)[name = tensor<string, []>("op_523_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_524_cast_fp16 = softmax(axis = var_384, x = aw_41_cast_fp16)[name = tensor<string, []>("op_524_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_525_cast_fp16 = softmax(axis = var_384, x = aw_43_cast_fp16)[name = tensor<string, []>("op_525_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_526_cast_fp16 = softmax(axis = var_384, x = aw_45_cast_fp16)[name = tensor<string, []>("op_526_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_527_cast_fp16 = softmax(axis = var_384, x = aw_47_cast_fp16)[name = tensor<string, []>("op_527_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_528_cast_fp16 = softmax(axis = var_384, x = aw_49_cast_fp16)[name = tensor<string, []>("op_528_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_529_cast_fp16 = softmax(axis = var_384, x = aw_51_cast_fp16)[name = tensor<string, []>("op_529_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_530_cast_fp16 = softmax(axis = var_384, x = aw_53_cast_fp16)[name = tensor<string, []>("op_530_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_531_cast_fp16 = softmax(axis = var_384, x = aw_55_cast_fp16)[name = tensor<string, []>("op_531_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_532_cast_fp16 = softmax(axis = var_384, x = aw_57_cast_fp16)[name = tensor<string, []>("op_532_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_533_cast_fp16 = softmax(axis = var_384, x = aw_59_cast_fp16)[name = tensor<string, []>("op_533_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_534_cast_fp16 = softmax(axis = var_384, x = aw_61_cast_fp16)[name = tensor<string, []>("op_534_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_535_cast_fp16 = softmax(axis = var_384, x = aw_63_cast_fp16)[name = tensor<string, []>("op_535_cast_fp16")];
            tensor<string, []> var_537_equation_0 = const()[name = tensor<string, []>("op_537_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_537_cast_fp16 = einsum(equation = var_537_equation_0, values = (var_471_cast_fp16_0, var_520_cast_fp16))[name = tensor<string, []>("op_537_cast_fp16")];
            tensor<string, []> var_539_equation_0 = const()[name = tensor<string, []>("op_539_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_539_cast_fp16 = einsum(equation = var_539_equation_0, values = (var_471_cast_fp16_1, var_521_cast_fp16))[name = tensor<string, []>("op_539_cast_fp16")];
            tensor<string, []> var_541_equation_0 = const()[name = tensor<string, []>("op_541_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_541_cast_fp16 = einsum(equation = var_541_equation_0, values = (var_471_cast_fp16_2, var_522_cast_fp16))[name = tensor<string, []>("op_541_cast_fp16")];
            tensor<string, []> var_543_equation_0 = const()[name = tensor<string, []>("op_543_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_543_cast_fp16 = einsum(equation = var_543_equation_0, values = (var_471_cast_fp16_3, var_523_cast_fp16))[name = tensor<string, []>("op_543_cast_fp16")];
            tensor<string, []> var_545_equation_0 = const()[name = tensor<string, []>("op_545_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_545_cast_fp16 = einsum(equation = var_545_equation_0, values = (var_471_cast_fp16_4, var_524_cast_fp16))[name = tensor<string, []>("op_545_cast_fp16")];
            tensor<string, []> var_547_equation_0 = const()[name = tensor<string, []>("op_547_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_547_cast_fp16 = einsum(equation = var_547_equation_0, values = (var_471_cast_fp16_5, var_525_cast_fp16))[name = tensor<string, []>("op_547_cast_fp16")];
            tensor<string, []> var_549_equation_0 = const()[name = tensor<string, []>("op_549_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_549_cast_fp16 = einsum(equation = var_549_equation_0, values = (var_471_cast_fp16_6, var_526_cast_fp16))[name = tensor<string, []>("op_549_cast_fp16")];
            tensor<string, []> var_551_equation_0 = const()[name = tensor<string, []>("op_551_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_551_cast_fp16 = einsum(equation = var_551_equation_0, values = (var_471_cast_fp16_7, var_527_cast_fp16))[name = tensor<string, []>("op_551_cast_fp16")];
            tensor<string, []> var_553_equation_0 = const()[name = tensor<string, []>("op_553_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_553_cast_fp16 = einsum(equation = var_553_equation_0, values = (var_471_cast_fp16_8, var_528_cast_fp16))[name = tensor<string, []>("op_553_cast_fp16")];
            tensor<string, []> var_555_equation_0 = const()[name = tensor<string, []>("op_555_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_471_cast_fp16_9, var_529_cast_fp16))[name = tensor<string, []>("op_555_cast_fp16")];
            tensor<string, []> var_557_equation_0 = const()[name = tensor<string, []>("op_557_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_557_cast_fp16 = einsum(equation = var_557_equation_0, values = (var_471_cast_fp16_10, var_530_cast_fp16))[name = tensor<string, []>("op_557_cast_fp16")];
            tensor<string, []> var_559_equation_0 = const()[name = tensor<string, []>("op_559_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_471_cast_fp16_11, var_531_cast_fp16))[name = tensor<string, []>("op_559_cast_fp16")];
            tensor<string, []> var_561_equation_0 = const()[name = tensor<string, []>("op_561_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_561_cast_fp16 = einsum(equation = var_561_equation_0, values = (var_471_cast_fp16_12, var_532_cast_fp16))[name = tensor<string, []>("op_561_cast_fp16")];
            tensor<string, []> var_563_equation_0 = const()[name = tensor<string, []>("op_563_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_471_cast_fp16_13, var_533_cast_fp16))[name = tensor<string, []>("op_563_cast_fp16")];
            tensor<string, []> var_565_equation_0 = const()[name = tensor<string, []>("op_565_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_565_cast_fp16 = einsum(equation = var_565_equation_0, values = (var_471_cast_fp16_14, var_534_cast_fp16))[name = tensor<string, []>("op_565_cast_fp16")];
            tensor<string, []> var_567_equation_0 = const()[name = tensor<string, []>("op_567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_471_cast_fp16_15, var_535_cast_fp16))[name = tensor<string, []>("op_567_cast_fp16")];
            tensor<bool, []> input_17_interleave_0 = const()[name = tensor<string, []>("input_17_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_17_cast_fp16 = concat(axis = var_384, interleave = input_17_interleave_0, values = (var_537_cast_fp16, var_539_cast_fp16, var_541_cast_fp16, var_543_cast_fp16, var_545_cast_fp16, var_547_cast_fp16, var_549_cast_fp16, var_551_cast_fp16, var_553_cast_fp16, var_555_cast_fp16, var_557_cast_fp16, var_559_cast_fp16, var_561_cast_fp16, var_563_cast_fp16, var_565_cast_fp16, var_567_cast_fp16))[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<string, []> var_576_pad_type_0 = const()[name = tensor<string, []>("op_576_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_576_strides_0 = const()[name = tensor<string, []>("op_576_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_576_pad_0 = const()[name = tensor<string, []>("op_576_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_576_dilations_0 = const()[name = tensor<string, []>("op_576_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_576_groups_0 = const()[name = tensor<string, []>("op_576_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_1_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9907456))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10431808))), name = tensor<string, []>("blocks_1_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_1_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10431936)))];
            tensor<fp16, [1, 1024, 1, 600]> var_576_cast_fp16 = conv(bias = blocks_1_attn_out_bias_to_fp16, dilations = var_576_dilations_0, groups = var_576_groups_0, pad = var_576_pad_0, pad_type = var_576_pad_type_0, strides = var_576_strides_0, weight = blocks_1_attn_out_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("op_576_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = var_576_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")];
            tensor<int32, [1]> input_19_axes_0 = const()[name = tensor<string, []>("input_19_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_19_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_19_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10434048)))];
            tensor<fp16, [1024]> input_19_beta_0_to_fp16 = const()[name = tensor<string, []>("input_19_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10436160)))];
            tensor<fp16, []> var_586_to_fp16 = const()[name = tensor<string, []>("op_586_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_19_cast_fp16 = layer_norm(axes = input_19_axes_0, beta = input_19_beta_0_to_fp16, epsilon = var_586_to_fp16, gamma = input_19_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<string, []> input_21_pad_type_0 = const()[name = tensor<string, []>("input_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_21_strides_0 = const()[name = tensor<string, []>("input_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_21_pad_0 = const()[name = tensor<string, []>("input_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_21_dilations_0 = const()[name = tensor<string, []>("input_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_21_groups_0 = const()[name = tensor<string, []>("input_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_1_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10438272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12535488))), name = tensor<string, []>("blocks_1_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_1_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12535616)))];
            tensor<fp16, [1, 4096, 1, 600]> input_21_cast_fp16 = conv(bias = blocks_1_mlp_0_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = blocks_1_mlp_0_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<string, []> input_23_mode_0 = const()[name = tensor<string, []>("input_23_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")];
            tensor<string, []> var_612_pad_type_0 = const()[name = tensor<string, []>("op_612_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_612_strides_0 = const()[name = tensor<string, []>("op_612_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_612_pad_0 = const()[name = tensor<string, []>("op_612_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_612_dilations_0 = const()[name = tensor<string, []>("op_612_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_612_groups_0 = const()[name = tensor<string, []>("op_612_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_1_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12543872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14641088))), name = tensor<string, []>("blocks_1_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_1_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14641216)))];
            tensor<fp16, [1, 1024, 1, 600]> var_612_cast_fp16 = conv(bias = blocks_1_mlp_2_bias_to_fp16, dilations = var_612_dilations_0, groups = var_612_groups_0, pad = var_612_pad_0, pad_type = var_612_pad_type_0, strides = var_612_strides_0, weight = blocks_1_mlp_2_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor<string, []>("op_612_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_612_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")];
            tensor<int32, []> var_621 = const()[name = tensor<string, []>("op_621"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_25_axes_0 = const()[name = tensor<string, []>("input_25_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_25_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14643328)))];
            tensor<fp16, [1024]> input_25_beta_0_to_fp16 = const()[name = tensor<string, []>("input_25_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14645440)))];
            tensor<fp16, []> var_637_to_fp16 = const()[name = tensor<string, []>("op_637_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_25_cast_fp16 = layer_norm(axes = input_25_axes_0, beta = input_25_beta_0_to_fp16, epsilon = var_637_to_fp16, gamma = input_25_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<string, []> q_5_pad_type_0 = const()[name = tensor<string, []>("q_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_5_strides_0 = const()[name = tensor<string, []>("q_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_5_pad_0 = const()[name = tensor<string, []>("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_5_dilations_0 = const()[name = tensor<string, []>("q_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_5_groups_0 = const()[name = tensor<string, []>("q_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_672_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14647552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15171904))), name = tensor<string, []>("op_672_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_672_bias_0_to_fp16 = const()[name = tensor<string, []>("op_672_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15172032)))];
            tensor<fp16, [1, 1024, 1, 600]> var_672_cast_fp16 = conv(bias = var_672_bias_0_to_fp16, dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = op_672_weight_0_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("op_672_cast_fp16")];
            tensor<string, []> k_5_pad_type_0 = const()[name = tensor<string, []>("k_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_5_strides_0 = const()[name = tensor<string, []>("k_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_5_pad_0 = const()[name = tensor<string, []>("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_5_dilations_0 = const()[name = tensor<string, []>("k_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_5_groups_0 = const()[name = tensor<string, []>("k_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_2_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15174144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15698496))), name = tensor<string, []>("blocks_2_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = blocks_2_attn_key_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")];
            tensor<string, []> var_670_pad_type_0 = const()[name = tensor<string, []>("op_670_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_670_strides_0 = const()[name = tensor<string, []>("op_670_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_670_pad_0 = const()[name = tensor<string, []>("op_670_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_670_dilations_0 = const()[name = tensor<string, []>("op_670_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_670_groups_0 = const()[name = tensor<string, []>("op_670_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_2_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15698624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16222976))), name = tensor<string, []>("blocks_2_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_2_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16223104)))];
            tensor<fp16, [1, 1024, 1, 600]> var_670_cast_fp16 = conv(bias = blocks_2_attn_value_bias_to_fp16, dilations = var_670_dilations_0, groups = var_670_groups_0, pad = var_670_pad_0, pad_type = var_670_pad_type_0, strides = var_670_strides_0, weight = blocks_2_attn_value_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("op_670_cast_fp16")];
            tensor<int32, [16]> tile_6 = const()[name = tensor<string, []>("tile_6"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_673_axis_0 = const()[name = tensor<string, []>("op_673_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_673_cast_fp16_15 = split(axis = var_673_axis_0, split_sizes = tile_6, x = var_672_cast_fp16)[name = tensor<string, []>("op_673_cast_fp16")];
            tensor<int32, [4]> var_690_perm_0 = const()[name = tensor<string, []>("op_690_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_7 = const()[name = tensor<string, []>("tile_7"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_691_axis_0 = const()[name = tensor<string, []>("op_691_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_690_cast_fp16 = transpose(perm = var_690_perm_0, x = k_5_cast_fp16)[name = tensor<string, []>("transpose_33")];
            tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_691_cast_fp16_15 = split(axis = var_691_axis_0, split_sizes = tile_7, x = var_690_cast_fp16)[name = tensor<string, []>("op_691_cast_fp16")];
            tensor<int32, [16]> tile_8 = const()[name = tensor<string, []>("tile_8"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_708_axis_0 = const()[name = tensor<string, []>("op_708_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_708_cast_fp16_15 = split(axis = var_708_axis_0, split_sizes = tile_8, x = var_670_cast_fp16)[name = tensor<string, []>("op_708_cast_fp16")];
            tensor<string, []> aw_65_equation_0 = const()[name = tensor<string, []>("aw_65_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_65_cast_fp16 = einsum(equation = aw_65_equation_0, values = (var_691_cast_fp16_0, var_673_cast_fp16_0))[name = tensor<string, []>("aw_65_cast_fp16")];
            tensor<string, []> aw_67_equation_0 = const()[name = tensor<string, []>("aw_67_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_67_cast_fp16 = einsum(equation = aw_67_equation_0, values = (var_691_cast_fp16_1, var_673_cast_fp16_1))[name = tensor<string, []>("aw_67_cast_fp16")];
            tensor<string, []> aw_69_equation_0 = const()[name = tensor<string, []>("aw_69_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_69_cast_fp16 = einsum(equation = aw_69_equation_0, values = (var_691_cast_fp16_2, var_673_cast_fp16_2))[name = tensor<string, []>("aw_69_cast_fp16")];
            tensor<string, []> aw_71_equation_0 = const()[name = tensor<string, []>("aw_71_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_71_cast_fp16 = einsum(equation = aw_71_equation_0, values = (var_691_cast_fp16_3, var_673_cast_fp16_3))[name = tensor<string, []>("aw_71_cast_fp16")];
            tensor<string, []> aw_73_equation_0 = const()[name = tensor<string, []>("aw_73_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_73_cast_fp16 = einsum(equation = aw_73_equation_0, values = (var_691_cast_fp16_4, var_673_cast_fp16_4))[name = tensor<string, []>("aw_73_cast_fp16")];
            tensor<string, []> aw_75_equation_0 = const()[name = tensor<string, []>("aw_75_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_75_cast_fp16 = einsum(equation = aw_75_equation_0, values = (var_691_cast_fp16_5, var_673_cast_fp16_5))[name = tensor<string, []>("aw_75_cast_fp16")];
            tensor<string, []> aw_77_equation_0 = const()[name = tensor<string, []>("aw_77_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_77_cast_fp16 = einsum(equation = aw_77_equation_0, values = (var_691_cast_fp16_6, var_673_cast_fp16_6))[name = tensor<string, []>("aw_77_cast_fp16")];
            tensor<string, []> aw_79_equation_0 = const()[name = tensor<string, []>("aw_79_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_79_cast_fp16 = einsum(equation = aw_79_equation_0, values = (var_691_cast_fp16_7, var_673_cast_fp16_7))[name = tensor<string, []>("aw_79_cast_fp16")];
            tensor<string, []> aw_81_equation_0 = const()[name = tensor<string, []>("aw_81_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_81_cast_fp16 = einsum(equation = aw_81_equation_0, values = (var_691_cast_fp16_8, var_673_cast_fp16_8))[name = tensor<string, []>("aw_81_cast_fp16")];
            tensor<string, []> aw_83_equation_0 = const()[name = tensor<string, []>("aw_83_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_83_cast_fp16 = einsum(equation = aw_83_equation_0, values = (var_691_cast_fp16_9, var_673_cast_fp16_9))[name = tensor<string, []>("aw_83_cast_fp16")];
            tensor<string, []> aw_85_equation_0 = const()[name = tensor<string, []>("aw_85_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_85_cast_fp16 = einsum(equation = aw_85_equation_0, values = (var_691_cast_fp16_10, var_673_cast_fp16_10))[name = tensor<string, []>("aw_85_cast_fp16")];
            tensor<string, []> aw_87_equation_0 = const()[name = tensor<string, []>("aw_87_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_87_cast_fp16 = einsum(equation = aw_87_equation_0, values = (var_691_cast_fp16_11, var_673_cast_fp16_11))[name = tensor<string, []>("aw_87_cast_fp16")];
            tensor<string, []> aw_89_equation_0 = const()[name = tensor<string, []>("aw_89_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_89_cast_fp16 = einsum(equation = aw_89_equation_0, values = (var_691_cast_fp16_12, var_673_cast_fp16_12))[name = tensor<string, []>("aw_89_cast_fp16")];
            tensor<string, []> aw_91_equation_0 = const()[name = tensor<string, []>("aw_91_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_91_cast_fp16 = einsum(equation = aw_91_equation_0, values = (var_691_cast_fp16_13, var_673_cast_fp16_13))[name = tensor<string, []>("aw_91_cast_fp16")];
            tensor<string, []> aw_93_equation_0 = const()[name = tensor<string, []>("aw_93_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_93_cast_fp16 = einsum(equation = aw_93_equation_0, values = (var_691_cast_fp16_14, var_673_cast_fp16_14))[name = tensor<string, []>("aw_93_cast_fp16")];
            tensor<string, []> aw_95_equation_0 = const()[name = tensor<string, []>("aw_95_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_95_cast_fp16 = einsum(equation = aw_95_equation_0, values = (var_691_cast_fp16_15, var_673_cast_fp16_15))[name = tensor<string, []>("aw_95_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_757_cast_fp16 = softmax(axis = var_621, x = aw_65_cast_fp16)[name = tensor<string, []>("op_757_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_758_cast_fp16 = softmax(axis = var_621, x = aw_67_cast_fp16)[name = tensor<string, []>("op_758_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_759_cast_fp16 = softmax(axis = var_621, x = aw_69_cast_fp16)[name = tensor<string, []>("op_759_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_760_cast_fp16 = softmax(axis = var_621, x = aw_71_cast_fp16)[name = tensor<string, []>("op_760_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_761_cast_fp16 = softmax(axis = var_621, x = aw_73_cast_fp16)[name = tensor<string, []>("op_761_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_762_cast_fp16 = softmax(axis = var_621, x = aw_75_cast_fp16)[name = tensor<string, []>("op_762_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_763_cast_fp16 = softmax(axis = var_621, x = aw_77_cast_fp16)[name = tensor<string, []>("op_763_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_764_cast_fp16 = softmax(axis = var_621, x = aw_79_cast_fp16)[name = tensor<string, []>("op_764_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_765_cast_fp16 = softmax(axis = var_621, x = aw_81_cast_fp16)[name = tensor<string, []>("op_765_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_766_cast_fp16 = softmax(axis = var_621, x = aw_83_cast_fp16)[name = tensor<string, []>("op_766_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_767_cast_fp16 = softmax(axis = var_621, x = aw_85_cast_fp16)[name = tensor<string, []>("op_767_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_768_cast_fp16 = softmax(axis = var_621, x = aw_87_cast_fp16)[name = tensor<string, []>("op_768_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_769_cast_fp16 = softmax(axis = var_621, x = aw_89_cast_fp16)[name = tensor<string, []>("op_769_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_770_cast_fp16 = softmax(axis = var_621, x = aw_91_cast_fp16)[name = tensor<string, []>("op_770_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_771_cast_fp16 = softmax(axis = var_621, x = aw_93_cast_fp16)[name = tensor<string, []>("op_771_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_772_cast_fp16 = softmax(axis = var_621, x = aw_95_cast_fp16)[name = tensor<string, []>("op_772_cast_fp16")];
            tensor<string, []> var_774_equation_0 = const()[name = tensor<string, []>("op_774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_774_cast_fp16 = einsum(equation = var_774_equation_0, values = (var_708_cast_fp16_0, var_757_cast_fp16))[name = tensor<string, []>("op_774_cast_fp16")];
            tensor<string, []> var_776_equation_0 = const()[name = tensor<string, []>("op_776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_776_cast_fp16 = einsum(equation = var_776_equation_0, values = (var_708_cast_fp16_1, var_758_cast_fp16))[name = tensor<string, []>("op_776_cast_fp16")];
            tensor<string, []> var_778_equation_0 = const()[name = tensor<string, []>("op_778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_778_cast_fp16 = einsum(equation = var_778_equation_0, values = (var_708_cast_fp16_2, var_759_cast_fp16))[name = tensor<string, []>("op_778_cast_fp16")];
            tensor<string, []> var_780_equation_0 = const()[name = tensor<string, []>("op_780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_780_cast_fp16 = einsum(equation = var_780_equation_0, values = (var_708_cast_fp16_3, var_760_cast_fp16))[name = tensor<string, []>("op_780_cast_fp16")];
            tensor<string, []> var_782_equation_0 = const()[name = tensor<string, []>("op_782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_782_cast_fp16 = einsum(equation = var_782_equation_0, values = (var_708_cast_fp16_4, var_761_cast_fp16))[name = tensor<string, []>("op_782_cast_fp16")];
            tensor<string, []> var_784_equation_0 = const()[name = tensor<string, []>("op_784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_784_cast_fp16 = einsum(equation = var_784_equation_0, values = (var_708_cast_fp16_5, var_762_cast_fp16))[name = tensor<string, []>("op_784_cast_fp16")];
            tensor<string, []> var_786_equation_0 = const()[name = tensor<string, []>("op_786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_786_cast_fp16 = einsum(equation = var_786_equation_0, values = (var_708_cast_fp16_6, var_763_cast_fp16))[name = tensor<string, []>("op_786_cast_fp16")];
            tensor<string, []> var_788_equation_0 = const()[name = tensor<string, []>("op_788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_788_cast_fp16 = einsum(equation = var_788_equation_0, values = (var_708_cast_fp16_7, var_764_cast_fp16))[name = tensor<string, []>("op_788_cast_fp16")];
            tensor<string, []> var_790_equation_0 = const()[name = tensor<string, []>("op_790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_790_cast_fp16 = einsum(equation = var_790_equation_0, values = (var_708_cast_fp16_8, var_765_cast_fp16))[name = tensor<string, []>("op_790_cast_fp16")];
            tensor<string, []> var_792_equation_0 = const()[name = tensor<string, []>("op_792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_792_cast_fp16 = einsum(equation = var_792_equation_0, values = (var_708_cast_fp16_9, var_766_cast_fp16))[name = tensor<string, []>("op_792_cast_fp16")];
            tensor<string, []> var_794_equation_0 = const()[name = tensor<string, []>("op_794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_794_cast_fp16 = einsum(equation = var_794_equation_0, values = (var_708_cast_fp16_10, var_767_cast_fp16))[name = tensor<string, []>("op_794_cast_fp16")];
            tensor<string, []> var_796_equation_0 = const()[name = tensor<string, []>("op_796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_796_cast_fp16 = einsum(equation = var_796_equation_0, values = (var_708_cast_fp16_11, var_768_cast_fp16))[name = tensor<string, []>("op_796_cast_fp16")];
            tensor<string, []> var_798_equation_0 = const()[name = tensor<string, []>("op_798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_798_cast_fp16 = einsum(equation = var_798_equation_0, values = (var_708_cast_fp16_12, var_769_cast_fp16))[name = tensor<string, []>("op_798_cast_fp16")];
            tensor<string, []> var_800_equation_0 = const()[name = tensor<string, []>("op_800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_800_cast_fp16 = einsum(equation = var_800_equation_0, values = (var_708_cast_fp16_13, var_770_cast_fp16))[name = tensor<string, []>("op_800_cast_fp16")];
            tensor<string, []> var_802_equation_0 = const()[name = tensor<string, []>("op_802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_802_cast_fp16 = einsum(equation = var_802_equation_0, values = (var_708_cast_fp16_14, var_771_cast_fp16))[name = tensor<string, []>("op_802_cast_fp16")];
            tensor<string, []> var_804_equation_0 = const()[name = tensor<string, []>("op_804_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_804_cast_fp16 = einsum(equation = var_804_equation_0, values = (var_708_cast_fp16_15, var_772_cast_fp16))[name = tensor<string, []>("op_804_cast_fp16")];
            tensor<bool, []> input_27_interleave_0 = const()[name = tensor<string, []>("input_27_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_27_cast_fp16 = concat(axis = var_621, interleave = input_27_interleave_0, values = (var_774_cast_fp16, var_776_cast_fp16, var_778_cast_fp16, var_780_cast_fp16, var_782_cast_fp16, var_784_cast_fp16, var_786_cast_fp16, var_788_cast_fp16, var_790_cast_fp16, var_792_cast_fp16, var_794_cast_fp16, var_796_cast_fp16, var_798_cast_fp16, var_800_cast_fp16, var_802_cast_fp16, var_804_cast_fp16))[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<string, []> var_813_pad_type_0 = const()[name = tensor<string, []>("op_813_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_813_strides_0 = const()[name = tensor<string, []>("op_813_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_813_pad_0 = const()[name = tensor<string, []>("op_813_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_813_dilations_0 = const()[name = tensor<string, []>("op_813_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_813_groups_0 = const()[name = tensor<string, []>("op_813_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_2_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16225216))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16749568))), name = tensor<string, []>("blocks_2_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_2_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16749696)))];
            tensor<fp16, [1, 1024, 1, 600]> var_813_cast_fp16 = conv(bias = blocks_2_attn_out_bias_to_fp16, dilations = var_813_dilations_0, groups = var_813_groups_0, pad = var_813_pad_0, pad_type = var_813_pad_type_0, strides = var_813_strides_0, weight = blocks_2_attn_out_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor<string, []>("op_813_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = var_813_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")];
            tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_29_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16751808)))];
            tensor<fp16, [1024]> input_29_beta_0_to_fp16 = const()[name = tensor<string, []>("input_29_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16753920)))];
            tensor<fp16, []> var_823_to_fp16 = const()[name = tensor<string, []>("op_823_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = input_29_beta_0_to_fp16, epsilon = var_823_to_fp16, gamma = input_29_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")];
            tensor<string, []> input_31_pad_type_0 = const()[name = tensor<string, []>("input_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_31_strides_0 = const()[name = tensor<string, []>("input_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_31_pad_0 = const()[name = tensor<string, []>("input_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_31_dilations_0 = const()[name = tensor<string, []>("input_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_31_groups_0 = const()[name = tensor<string, []>("input_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_2_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16756032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18853248))), name = tensor<string, []>("blocks_2_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_2_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18853376)))];
            tensor<fp16, [1, 4096, 1, 600]> input_31_cast_fp16 = conv(bias = blocks_2_mlp_0_bias_to_fp16, dilations = input_31_dilations_0, groups = input_31_groups_0, pad = input_31_pad_0, pad_type = input_31_pad_type_0, strides = input_31_strides_0, weight = blocks_2_mlp_0_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_33_cast_fp16 = gelu(mode = input_33_mode_0, x = input_31_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")];
            tensor<string, []> var_849_pad_type_0 = const()[name = tensor<string, []>("op_849_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_849_strides_0 = const()[name = tensor<string, []>("op_849_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_849_pad_0 = const()[name = tensor<string, []>("op_849_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_849_dilations_0 = const()[name = tensor<string, []>("op_849_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_849_groups_0 = const()[name = tensor<string, []>("op_849_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_2_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18861632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20958848))), name = tensor<string, []>("blocks_2_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_2_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20958976)))];
            tensor<fp16, [1, 1024, 1, 600]> var_849_cast_fp16 = conv(bias = blocks_2_mlp_2_bias_to_fp16, dilations = var_849_dilations_0, groups = var_849_groups_0, pad = var_849_pad_0, pad_type = var_849_pad_type_0, strides = var_849_strides_0, weight = blocks_2_mlp_2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor<string, []>("op_849_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_849_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")];
            tensor<int32, []> var_858 = const()[name = tensor<string, []>("op_858"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_35_axes_0 = const()[name = tensor<string, []>("input_35_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20961088)))];
            tensor<fp16, [1024]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20963200)))];
            tensor<fp16, []> var_874_to_fp16 = const()[name = tensor<string, []>("op_874_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_35_cast_fp16 = layer_norm(axes = input_35_axes_0, beta = input_35_beta_0_to_fp16, epsilon = var_874_to_fp16, gamma = input_35_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<string, []> q_7_pad_type_0 = const()[name = tensor<string, []>("q_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_7_strides_0 = const()[name = tensor<string, []>("q_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_7_pad_0 = const()[name = tensor<string, []>("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_7_dilations_0 = const()[name = tensor<string, []>("q_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_7_groups_0 = const()[name = tensor<string, []>("q_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_909_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20965312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21489664))), name = tensor<string, []>("op_909_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_909_bias_0_to_fp16 = const()[name = tensor<string, []>("op_909_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21489792)))];
            tensor<fp16, [1, 1024, 1, 600]> var_909_cast_fp16 = conv(bias = var_909_bias_0_to_fp16, dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = op_909_weight_0_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("op_909_cast_fp16")];
            tensor<string, []> k_7_pad_type_0 = const()[name = tensor<string, []>("k_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_7_strides_0 = const()[name = tensor<string, []>("k_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_7_pad_0 = const()[name = tensor<string, []>("k_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_7_dilations_0 = const()[name = tensor<string, []>("k_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_7_groups_0 = const()[name = tensor<string, []>("k_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_3_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21491904))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22016256))), name = tensor<string, []>("blocks_3_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_7_cast_fp16 = conv(dilations = k_7_dilations_0, groups = k_7_groups_0, pad = k_7_pad_0, pad_type = k_7_pad_type_0, strides = k_7_strides_0, weight = blocks_3_attn_key_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("k_7_cast_fp16")];
            tensor<string, []> var_907_pad_type_0 = const()[name = tensor<string, []>("op_907_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_907_strides_0 = const()[name = tensor<string, []>("op_907_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_907_pad_0 = const()[name = tensor<string, []>("op_907_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_907_dilations_0 = const()[name = tensor<string, []>("op_907_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_907_groups_0 = const()[name = tensor<string, []>("op_907_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_3_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22016384))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22540736))), name = tensor<string, []>("blocks_3_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_3_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22540864)))];
            tensor<fp16, [1, 1024, 1, 600]> var_907_cast_fp16 = conv(bias = blocks_3_attn_value_bias_to_fp16, dilations = var_907_dilations_0, groups = var_907_groups_0, pad = var_907_pad_0, pad_type = var_907_pad_type_0, strides = var_907_strides_0, weight = blocks_3_attn_value_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("op_907_cast_fp16")];
            tensor<int32, [16]> tile_9 = const()[name = tensor<string, []>("tile_9"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_910_axis_0 = const()[name = tensor<string, []>("op_910_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_910_cast_fp16_15 = split(axis = var_910_axis_0, split_sizes = tile_9, x = var_909_cast_fp16)[name = tensor<string, []>("op_910_cast_fp16")];
            tensor<int32, [4]> var_927_perm_0 = const()[name = tensor<string, []>("op_927_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_10 = const()[name = tensor<string, []>("tile_10"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_928_axis_0 = const()[name = tensor<string, []>("op_928_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_927_cast_fp16 = transpose(perm = var_927_perm_0, x = k_7_cast_fp16)[name = tensor<string, []>("transpose_32")];
            tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_928_cast_fp16_15 = split(axis = var_928_axis_0, split_sizes = tile_10, x = var_927_cast_fp16)[name = tensor<string, []>("op_928_cast_fp16")];
            tensor<int32, [16]> tile_11 = const()[name = tensor<string, []>("tile_11"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_945_axis_0 = const()[name = tensor<string, []>("op_945_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_945_cast_fp16_15 = split(axis = var_945_axis_0, split_sizes = tile_11, x = var_907_cast_fp16)[name = tensor<string, []>("op_945_cast_fp16")];
            tensor<string, []> aw_97_equation_0 = const()[name = tensor<string, []>("aw_97_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_97_cast_fp16 = einsum(equation = aw_97_equation_0, values = (var_928_cast_fp16_0, var_910_cast_fp16_0))[name = tensor<string, []>("aw_97_cast_fp16")];
            tensor<string, []> aw_99_equation_0 = const()[name = tensor<string, []>("aw_99_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_99_cast_fp16 = einsum(equation = aw_99_equation_0, values = (var_928_cast_fp16_1, var_910_cast_fp16_1))[name = tensor<string, []>("aw_99_cast_fp16")];
            tensor<string, []> aw_101_equation_0 = const()[name = tensor<string, []>("aw_101_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_101_cast_fp16 = einsum(equation = aw_101_equation_0, values = (var_928_cast_fp16_2, var_910_cast_fp16_2))[name = tensor<string, []>("aw_101_cast_fp16")];
            tensor<string, []> aw_103_equation_0 = const()[name = tensor<string, []>("aw_103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_103_cast_fp16 = einsum(equation = aw_103_equation_0, values = (var_928_cast_fp16_3, var_910_cast_fp16_3))[name = tensor<string, []>("aw_103_cast_fp16")];
            tensor<string, []> aw_105_equation_0 = const()[name = tensor<string, []>("aw_105_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_105_cast_fp16 = einsum(equation = aw_105_equation_0, values = (var_928_cast_fp16_4, var_910_cast_fp16_4))[name = tensor<string, []>("aw_105_cast_fp16")];
            tensor<string, []> aw_107_equation_0 = const()[name = tensor<string, []>("aw_107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_107_cast_fp16 = einsum(equation = aw_107_equation_0, values = (var_928_cast_fp16_5, var_910_cast_fp16_5))[name = tensor<string, []>("aw_107_cast_fp16")];
            tensor<string, []> aw_109_equation_0 = const()[name = tensor<string, []>("aw_109_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_109_cast_fp16 = einsum(equation = aw_109_equation_0, values = (var_928_cast_fp16_6, var_910_cast_fp16_6))[name = tensor<string, []>("aw_109_cast_fp16")];
            tensor<string, []> aw_111_equation_0 = const()[name = tensor<string, []>("aw_111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_111_cast_fp16 = einsum(equation = aw_111_equation_0, values = (var_928_cast_fp16_7, var_910_cast_fp16_7))[name = tensor<string, []>("aw_111_cast_fp16")];
            tensor<string, []> aw_113_equation_0 = const()[name = tensor<string, []>("aw_113_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_113_cast_fp16 = einsum(equation = aw_113_equation_0, values = (var_928_cast_fp16_8, var_910_cast_fp16_8))[name = tensor<string, []>("aw_113_cast_fp16")];
            tensor<string, []> aw_115_equation_0 = const()[name = tensor<string, []>("aw_115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_115_cast_fp16 = einsum(equation = aw_115_equation_0, values = (var_928_cast_fp16_9, var_910_cast_fp16_9))[name = tensor<string, []>("aw_115_cast_fp16")];
            tensor<string, []> aw_117_equation_0 = const()[name = tensor<string, []>("aw_117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_117_cast_fp16 = einsum(equation = aw_117_equation_0, values = (var_928_cast_fp16_10, var_910_cast_fp16_10))[name = tensor<string, []>("aw_117_cast_fp16")];
            tensor<string, []> aw_119_equation_0 = const()[name = tensor<string, []>("aw_119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_119_cast_fp16 = einsum(equation = aw_119_equation_0, values = (var_928_cast_fp16_11, var_910_cast_fp16_11))[name = tensor<string, []>("aw_119_cast_fp16")];
            tensor<string, []> aw_121_equation_0 = const()[name = tensor<string, []>("aw_121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_121_cast_fp16 = einsum(equation = aw_121_equation_0, values = (var_928_cast_fp16_12, var_910_cast_fp16_12))[name = tensor<string, []>("aw_121_cast_fp16")];
            tensor<string, []> aw_123_equation_0 = const()[name = tensor<string, []>("aw_123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_123_cast_fp16 = einsum(equation = aw_123_equation_0, values = (var_928_cast_fp16_13, var_910_cast_fp16_13))[name = tensor<string, []>("aw_123_cast_fp16")];
            tensor<string, []> aw_125_equation_0 = const()[name = tensor<string, []>("aw_125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_125_cast_fp16 = einsum(equation = aw_125_equation_0, values = (var_928_cast_fp16_14, var_910_cast_fp16_14))[name = tensor<string, []>("aw_125_cast_fp16")];
            tensor<string, []> aw_127_equation_0 = const()[name = tensor<string, []>("aw_127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_127_cast_fp16 = einsum(equation = aw_127_equation_0, values = (var_928_cast_fp16_15, var_910_cast_fp16_15))[name = tensor<string, []>("aw_127_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_994_cast_fp16 = softmax(axis = var_858, x = aw_97_cast_fp16)[name = tensor<string, []>("op_994_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_995_cast_fp16 = softmax(axis = var_858, x = aw_99_cast_fp16)[name = tensor<string, []>("op_995_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_996_cast_fp16 = softmax(axis = var_858, x = aw_101_cast_fp16)[name = tensor<string, []>("op_996_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_997_cast_fp16 = softmax(axis = var_858, x = aw_103_cast_fp16)[name = tensor<string, []>("op_997_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_998_cast_fp16 = softmax(axis = var_858, x = aw_105_cast_fp16)[name = tensor<string, []>("op_998_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_999_cast_fp16 = softmax(axis = var_858, x = aw_107_cast_fp16)[name = tensor<string, []>("op_999_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1000_cast_fp16 = softmax(axis = var_858, x = aw_109_cast_fp16)[name = tensor<string, []>("op_1000_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1001_cast_fp16 = softmax(axis = var_858, x = aw_111_cast_fp16)[name = tensor<string, []>("op_1001_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1002_cast_fp16 = softmax(axis = var_858, x = aw_113_cast_fp16)[name = tensor<string, []>("op_1002_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1003_cast_fp16 = softmax(axis = var_858, x = aw_115_cast_fp16)[name = tensor<string, []>("op_1003_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1004_cast_fp16 = softmax(axis = var_858, x = aw_117_cast_fp16)[name = tensor<string, []>("op_1004_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1005_cast_fp16 = softmax(axis = var_858, x = aw_119_cast_fp16)[name = tensor<string, []>("op_1005_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1006_cast_fp16 = softmax(axis = var_858, x = aw_121_cast_fp16)[name = tensor<string, []>("op_1006_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1007_cast_fp16 = softmax(axis = var_858, x = aw_123_cast_fp16)[name = tensor<string, []>("op_1007_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1008_cast_fp16 = softmax(axis = var_858, x = aw_125_cast_fp16)[name = tensor<string, []>("op_1008_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1009_cast_fp16 = softmax(axis = var_858, x = aw_127_cast_fp16)[name = tensor<string, []>("op_1009_cast_fp16")];
            tensor<string, []> var_1011_equation_0 = const()[name = tensor<string, []>("op_1011_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1011_cast_fp16 = einsum(equation = var_1011_equation_0, values = (var_945_cast_fp16_0, var_994_cast_fp16))[name = tensor<string, []>("op_1011_cast_fp16")];
            tensor<string, []> var_1013_equation_0 = const()[name = tensor<string, []>("op_1013_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1013_cast_fp16 = einsum(equation = var_1013_equation_0, values = (var_945_cast_fp16_1, var_995_cast_fp16))[name = tensor<string, []>("op_1013_cast_fp16")];
            tensor<string, []> var_1015_equation_0 = const()[name = tensor<string, []>("op_1015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1015_cast_fp16 = einsum(equation = var_1015_equation_0, values = (var_945_cast_fp16_2, var_996_cast_fp16))[name = tensor<string, []>("op_1015_cast_fp16")];
            tensor<string, []> var_1017_equation_0 = const()[name = tensor<string, []>("op_1017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1017_cast_fp16 = einsum(equation = var_1017_equation_0, values = (var_945_cast_fp16_3, var_997_cast_fp16))[name = tensor<string, []>("op_1017_cast_fp16")];
            tensor<string, []> var_1019_equation_0 = const()[name = tensor<string, []>("op_1019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1019_cast_fp16 = einsum(equation = var_1019_equation_0, values = (var_945_cast_fp16_4, var_998_cast_fp16))[name = tensor<string, []>("op_1019_cast_fp16")];
            tensor<string, []> var_1021_equation_0 = const()[name = tensor<string, []>("op_1021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1021_cast_fp16 = einsum(equation = var_1021_equation_0, values = (var_945_cast_fp16_5, var_999_cast_fp16))[name = tensor<string, []>("op_1021_cast_fp16")];
            tensor<string, []> var_1023_equation_0 = const()[name = tensor<string, []>("op_1023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1023_cast_fp16 = einsum(equation = var_1023_equation_0, values = (var_945_cast_fp16_6, var_1000_cast_fp16))[name = tensor<string, []>("op_1023_cast_fp16")];
            tensor<string, []> var_1025_equation_0 = const()[name = tensor<string, []>("op_1025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1025_cast_fp16 = einsum(equation = var_1025_equation_0, values = (var_945_cast_fp16_7, var_1001_cast_fp16))[name = tensor<string, []>("op_1025_cast_fp16")];
            tensor<string, []> var_1027_equation_0 = const()[name = tensor<string, []>("op_1027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1027_cast_fp16 = einsum(equation = var_1027_equation_0, values = (var_945_cast_fp16_8, var_1002_cast_fp16))[name = tensor<string, []>("op_1027_cast_fp16")];
            tensor<string, []> var_1029_equation_0 = const()[name = tensor<string, []>("op_1029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1029_cast_fp16 = einsum(equation = var_1029_equation_0, values = (var_945_cast_fp16_9, var_1003_cast_fp16))[name = tensor<string, []>("op_1029_cast_fp16")];
            tensor<string, []> var_1031_equation_0 = const()[name = tensor<string, []>("op_1031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1031_cast_fp16 = einsum(equation = var_1031_equation_0, values = (var_945_cast_fp16_10, var_1004_cast_fp16))[name = tensor<string, []>("op_1031_cast_fp16")];
            tensor<string, []> var_1033_equation_0 = const()[name = tensor<string, []>("op_1033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1033_cast_fp16 = einsum(equation = var_1033_equation_0, values = (var_945_cast_fp16_11, var_1005_cast_fp16))[name = tensor<string, []>("op_1033_cast_fp16")];
            tensor<string, []> var_1035_equation_0 = const()[name = tensor<string, []>("op_1035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1035_cast_fp16 = einsum(equation = var_1035_equation_0, values = (var_945_cast_fp16_12, var_1006_cast_fp16))[name = tensor<string, []>("op_1035_cast_fp16")];
            tensor<string, []> var_1037_equation_0 = const()[name = tensor<string, []>("op_1037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1037_cast_fp16 = einsum(equation = var_1037_equation_0, values = (var_945_cast_fp16_13, var_1007_cast_fp16))[name = tensor<string, []>("op_1037_cast_fp16")];
            tensor<string, []> var_1039_equation_0 = const()[name = tensor<string, []>("op_1039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1039_cast_fp16 = einsum(equation = var_1039_equation_0, values = (var_945_cast_fp16_14, var_1008_cast_fp16))[name = tensor<string, []>("op_1039_cast_fp16")];
            tensor<string, []> var_1041_equation_0 = const()[name = tensor<string, []>("op_1041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1041_cast_fp16 = einsum(equation = var_1041_equation_0, values = (var_945_cast_fp16_15, var_1009_cast_fp16))[name = tensor<string, []>("op_1041_cast_fp16")];
            tensor<bool, []> input_37_interleave_0 = const()[name = tensor<string, []>("input_37_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_37_cast_fp16 = concat(axis = var_858, interleave = input_37_interleave_0, values = (var_1011_cast_fp16, var_1013_cast_fp16, var_1015_cast_fp16, var_1017_cast_fp16, var_1019_cast_fp16, var_1021_cast_fp16, var_1023_cast_fp16, var_1025_cast_fp16, var_1027_cast_fp16, var_1029_cast_fp16, var_1031_cast_fp16, var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16, var_1039_cast_fp16, var_1041_cast_fp16))[name = tensor<string, []>("input_37_cast_fp16")];
            tensor<string, []> var_1050_pad_type_0 = const()[name = tensor<string, []>("op_1050_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1050_strides_0 = const()[name = tensor<string, []>("op_1050_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1050_pad_0 = const()[name = tensor<string, []>("op_1050_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1050_dilations_0 = const()[name = tensor<string, []>("op_1050_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1050_groups_0 = const()[name = tensor<string, []>("op_1050_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_3_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22542976))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23067328))), name = tensor<string, []>("blocks_3_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_3_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23067456)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1050_cast_fp16 = conv(bias = blocks_3_attn_out_bias_to_fp16, dilations = var_1050_dilations_0, groups = var_1050_groups_0, pad = var_1050_pad_0, pad_type = var_1050_pad_type_0, strides = var_1050_strides_0, weight = blocks_3_attn_out_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor<string, []>("op_1050_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = var_1050_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")];
            tensor<int32, [1]> input_39_axes_0 = const()[name = tensor<string, []>("input_39_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_39_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_39_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23069568)))];
            tensor<fp16, [1024]> input_39_beta_0_to_fp16 = const()[name = tensor<string, []>("input_39_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23071680)))];
            tensor<fp16, []> var_1060_to_fp16 = const()[name = tensor<string, []>("op_1060_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_39_cast_fp16 = layer_norm(axes = input_39_axes_0, beta = input_39_beta_0_to_fp16, epsilon = var_1060_to_fp16, gamma = input_39_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
            tensor<string, []> input_41_pad_type_0 = const()[name = tensor<string, []>("input_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_41_strides_0 = const()[name = tensor<string, []>("input_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_41_pad_0 = const()[name = tensor<string, []>("input_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_41_dilations_0 = const()[name = tensor<string, []>("input_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_41_groups_0 = const()[name = tensor<string, []>("input_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_3_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23073792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25171008))), name = tensor<string, []>("blocks_3_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_3_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25171136)))];
            tensor<fp16, [1, 4096, 1, 600]> input_41_cast_fp16 = conv(bias = blocks_3_mlp_0_bias_to_fp16, dilations = input_41_dilations_0, groups = input_41_groups_0, pad = input_41_pad_0, pad_type = input_41_pad_type_0, strides = input_41_strides_0, weight = blocks_3_mlp_0_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<string, []> input_43_mode_0 = const()[name = tensor<string, []>("input_43_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_43_cast_fp16 = gelu(mode = input_43_mode_0, x = input_41_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<string, []> var_1086_pad_type_0 = const()[name = tensor<string, []>("op_1086_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1086_strides_0 = const()[name = tensor<string, []>("op_1086_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1086_pad_0 = const()[name = tensor<string, []>("op_1086_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1086_dilations_0 = const()[name = tensor<string, []>("op_1086_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1086_groups_0 = const()[name = tensor<string, []>("op_1086_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_3_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25179392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27276608))), name = tensor<string, []>("blocks_3_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_3_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27276736)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1086_cast_fp16 = conv(bias = blocks_3_mlp_2_bias_to_fp16, dilations = var_1086_dilations_0, groups = var_1086_groups_0, pad = var_1086_pad_0, pad_type = var_1086_pad_type_0, strides = var_1086_strides_0, weight = blocks_3_mlp_2_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor<string, []>("op_1086_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = var_1086_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")];
            tensor<int32, []> var_1095 = const()[name = tensor<string, []>("op_1095"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_45_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27278848)))];
            tensor<fp16, [1024]> input_45_beta_0_to_fp16 = const()[name = tensor<string, []>("input_45_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27280960)))];
            tensor<fp16, []> var_1111_to_fp16 = const()[name = tensor<string, []>("op_1111_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = input_45_beta_0_to_fp16, epsilon = var_1111_to_fp16, gamma = input_45_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<string, []> q_9_pad_type_0 = const()[name = tensor<string, []>("q_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_9_strides_0 = const()[name = tensor<string, []>("q_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_9_pad_0 = const()[name = tensor<string, []>("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_9_dilations_0 = const()[name = tensor<string, []>("q_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_9_groups_0 = const()[name = tensor<string, []>("q_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_1146_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27283072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27807424))), name = tensor<string, []>("op_1146_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_1146_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1146_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27807552)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1146_cast_fp16 = conv(bias = var_1146_bias_0_to_fp16, dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = op_1146_weight_0_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("op_1146_cast_fp16")];
            tensor<string, []> k_9_pad_type_0 = const()[name = tensor<string, []>("k_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_9_strides_0 = const()[name = tensor<string, []>("k_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_9_pad_0 = const()[name = tensor<string, []>("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_9_dilations_0 = const()[name = tensor<string, []>("k_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_9_groups_0 = const()[name = tensor<string, []>("k_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_4_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27809664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28334016))), name = tensor<string, []>("blocks_4_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = blocks_4_attn_key_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("k_9_cast_fp16")];
            tensor<string, []> var_1144_pad_type_0 = const()[name = tensor<string, []>("op_1144_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1144_strides_0 = const()[name = tensor<string, []>("op_1144_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1144_pad_0 = const()[name = tensor<string, []>("op_1144_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1144_dilations_0 = const()[name = tensor<string, []>("op_1144_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1144_groups_0 = const()[name = tensor<string, []>("op_1144_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_4_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28334144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28858496))), name = tensor<string, []>("blocks_4_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_4_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28858624)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1144_cast_fp16 = conv(bias = blocks_4_attn_value_bias_to_fp16, dilations = var_1144_dilations_0, groups = var_1144_groups_0, pad = var_1144_pad_0, pad_type = var_1144_pad_type_0, strides = var_1144_strides_0, weight = blocks_4_attn_value_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("op_1144_cast_fp16")];
            tensor<int32, [16]> tile_12 = const()[name = tensor<string, []>("tile_12"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1147_axis_0 = const()[name = tensor<string, []>("op_1147_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1147_cast_fp16_15 = split(axis = var_1147_axis_0, split_sizes = tile_12, x = var_1146_cast_fp16)[name = tensor<string, []>("op_1147_cast_fp16")];
            tensor<int32, [4]> var_1164_perm_0 = const()[name = tensor<string, []>("op_1164_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_13 = const()[name = tensor<string, []>("tile_13"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1165_axis_0 = const()[name = tensor<string, []>("op_1165_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_1164_cast_fp16 = transpose(perm = var_1164_perm_0, x = k_9_cast_fp16)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_1165_cast_fp16_15 = split(axis = var_1165_axis_0, split_sizes = tile_13, x = var_1164_cast_fp16)[name = tensor<string, []>("op_1165_cast_fp16")];
            tensor<int32, [16]> tile_14 = const()[name = tensor<string, []>("tile_14"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1182_axis_0 = const()[name = tensor<string, []>("op_1182_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1182_cast_fp16_15 = split(axis = var_1182_axis_0, split_sizes = tile_14, x = var_1144_cast_fp16)[name = tensor<string, []>("op_1182_cast_fp16")];
            tensor<string, []> aw_129_equation_0 = const()[name = tensor<string, []>("aw_129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_129_cast_fp16 = einsum(equation = aw_129_equation_0, values = (var_1165_cast_fp16_0, var_1147_cast_fp16_0))[name = tensor<string, []>("aw_129_cast_fp16")];
            tensor<string, []> aw_131_equation_0 = const()[name = tensor<string, []>("aw_131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_131_cast_fp16 = einsum(equation = aw_131_equation_0, values = (var_1165_cast_fp16_1, var_1147_cast_fp16_1))[name = tensor<string, []>("aw_131_cast_fp16")];
            tensor<string, []> aw_133_equation_0 = const()[name = tensor<string, []>("aw_133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_133_cast_fp16 = einsum(equation = aw_133_equation_0, values = (var_1165_cast_fp16_2, var_1147_cast_fp16_2))[name = tensor<string, []>("aw_133_cast_fp16")];
            tensor<string, []> aw_135_equation_0 = const()[name = tensor<string, []>("aw_135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_135_cast_fp16 = einsum(equation = aw_135_equation_0, values = (var_1165_cast_fp16_3, var_1147_cast_fp16_3))[name = tensor<string, []>("aw_135_cast_fp16")];
            tensor<string, []> aw_137_equation_0 = const()[name = tensor<string, []>("aw_137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_137_cast_fp16 = einsum(equation = aw_137_equation_0, values = (var_1165_cast_fp16_4, var_1147_cast_fp16_4))[name = tensor<string, []>("aw_137_cast_fp16")];
            tensor<string, []> aw_139_equation_0 = const()[name = tensor<string, []>("aw_139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_139_cast_fp16 = einsum(equation = aw_139_equation_0, values = (var_1165_cast_fp16_5, var_1147_cast_fp16_5))[name = tensor<string, []>("aw_139_cast_fp16")];
            tensor<string, []> aw_141_equation_0 = const()[name = tensor<string, []>("aw_141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_141_cast_fp16 = einsum(equation = aw_141_equation_0, values = (var_1165_cast_fp16_6, var_1147_cast_fp16_6))[name = tensor<string, []>("aw_141_cast_fp16")];
            tensor<string, []> aw_143_equation_0 = const()[name = tensor<string, []>("aw_143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_143_cast_fp16 = einsum(equation = aw_143_equation_0, values = (var_1165_cast_fp16_7, var_1147_cast_fp16_7))[name = tensor<string, []>("aw_143_cast_fp16")];
            tensor<string, []> aw_145_equation_0 = const()[name = tensor<string, []>("aw_145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_145_cast_fp16 = einsum(equation = aw_145_equation_0, values = (var_1165_cast_fp16_8, var_1147_cast_fp16_8))[name = tensor<string, []>("aw_145_cast_fp16")];
            tensor<string, []> aw_147_equation_0 = const()[name = tensor<string, []>("aw_147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_147_cast_fp16 = einsum(equation = aw_147_equation_0, values = (var_1165_cast_fp16_9, var_1147_cast_fp16_9))[name = tensor<string, []>("aw_147_cast_fp16")];
            tensor<string, []> aw_149_equation_0 = const()[name = tensor<string, []>("aw_149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_149_cast_fp16 = einsum(equation = aw_149_equation_0, values = (var_1165_cast_fp16_10, var_1147_cast_fp16_10))[name = tensor<string, []>("aw_149_cast_fp16")];
            tensor<string, []> aw_151_equation_0 = const()[name = tensor<string, []>("aw_151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_151_cast_fp16 = einsum(equation = aw_151_equation_0, values = (var_1165_cast_fp16_11, var_1147_cast_fp16_11))[name = tensor<string, []>("aw_151_cast_fp16")];
            tensor<string, []> aw_153_equation_0 = const()[name = tensor<string, []>("aw_153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_153_cast_fp16 = einsum(equation = aw_153_equation_0, values = (var_1165_cast_fp16_12, var_1147_cast_fp16_12))[name = tensor<string, []>("aw_153_cast_fp16")];
            tensor<string, []> aw_155_equation_0 = const()[name = tensor<string, []>("aw_155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_155_cast_fp16 = einsum(equation = aw_155_equation_0, values = (var_1165_cast_fp16_13, var_1147_cast_fp16_13))[name = tensor<string, []>("aw_155_cast_fp16")];
            tensor<string, []> aw_157_equation_0 = const()[name = tensor<string, []>("aw_157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_157_cast_fp16 = einsum(equation = aw_157_equation_0, values = (var_1165_cast_fp16_14, var_1147_cast_fp16_14))[name = tensor<string, []>("aw_157_cast_fp16")];
            tensor<string, []> aw_159_equation_0 = const()[name = tensor<string, []>("aw_159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_159_cast_fp16 = einsum(equation = aw_159_equation_0, values = (var_1165_cast_fp16_15, var_1147_cast_fp16_15))[name = tensor<string, []>("aw_159_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1231_cast_fp16 = softmax(axis = var_1095, x = aw_129_cast_fp16)[name = tensor<string, []>("op_1231_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1232_cast_fp16 = softmax(axis = var_1095, x = aw_131_cast_fp16)[name = tensor<string, []>("op_1232_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1233_cast_fp16 = softmax(axis = var_1095, x = aw_133_cast_fp16)[name = tensor<string, []>("op_1233_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1234_cast_fp16 = softmax(axis = var_1095, x = aw_135_cast_fp16)[name = tensor<string, []>("op_1234_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1235_cast_fp16 = softmax(axis = var_1095, x = aw_137_cast_fp16)[name = tensor<string, []>("op_1235_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1236_cast_fp16 = softmax(axis = var_1095, x = aw_139_cast_fp16)[name = tensor<string, []>("op_1236_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1237_cast_fp16 = softmax(axis = var_1095, x = aw_141_cast_fp16)[name = tensor<string, []>("op_1237_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1238_cast_fp16 = softmax(axis = var_1095, x = aw_143_cast_fp16)[name = tensor<string, []>("op_1238_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1239_cast_fp16 = softmax(axis = var_1095, x = aw_145_cast_fp16)[name = tensor<string, []>("op_1239_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1240_cast_fp16 = softmax(axis = var_1095, x = aw_147_cast_fp16)[name = tensor<string, []>("op_1240_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1241_cast_fp16 = softmax(axis = var_1095, x = aw_149_cast_fp16)[name = tensor<string, []>("op_1241_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1242_cast_fp16 = softmax(axis = var_1095, x = aw_151_cast_fp16)[name = tensor<string, []>("op_1242_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1243_cast_fp16 = softmax(axis = var_1095, x = aw_153_cast_fp16)[name = tensor<string, []>("op_1243_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1244_cast_fp16 = softmax(axis = var_1095, x = aw_155_cast_fp16)[name = tensor<string, []>("op_1244_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1245_cast_fp16 = softmax(axis = var_1095, x = aw_157_cast_fp16)[name = tensor<string, []>("op_1245_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1246_cast_fp16 = softmax(axis = var_1095, x = aw_159_cast_fp16)[name = tensor<string, []>("op_1246_cast_fp16")];
            tensor<string, []> var_1248_equation_0 = const()[name = tensor<string, []>("op_1248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1248_cast_fp16 = einsum(equation = var_1248_equation_0, values = (var_1182_cast_fp16_0, var_1231_cast_fp16))[name = tensor<string, []>("op_1248_cast_fp16")];
            tensor<string, []> var_1250_equation_0 = const()[name = tensor<string, []>("op_1250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1250_cast_fp16 = einsum(equation = var_1250_equation_0, values = (var_1182_cast_fp16_1, var_1232_cast_fp16))[name = tensor<string, []>("op_1250_cast_fp16")];
            tensor<string, []> var_1252_equation_0 = const()[name = tensor<string, []>("op_1252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1252_cast_fp16 = einsum(equation = var_1252_equation_0, values = (var_1182_cast_fp16_2, var_1233_cast_fp16))[name = tensor<string, []>("op_1252_cast_fp16")];
            tensor<string, []> var_1254_equation_0 = const()[name = tensor<string, []>("op_1254_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1254_cast_fp16 = einsum(equation = var_1254_equation_0, values = (var_1182_cast_fp16_3, var_1234_cast_fp16))[name = tensor<string, []>("op_1254_cast_fp16")];
            tensor<string, []> var_1256_equation_0 = const()[name = tensor<string, []>("op_1256_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1256_cast_fp16 = einsum(equation = var_1256_equation_0, values = (var_1182_cast_fp16_4, var_1235_cast_fp16))[name = tensor<string, []>("op_1256_cast_fp16")];
            tensor<string, []> var_1258_equation_0 = const()[name = tensor<string, []>("op_1258_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1258_cast_fp16 = einsum(equation = var_1258_equation_0, values = (var_1182_cast_fp16_5, var_1236_cast_fp16))[name = tensor<string, []>("op_1258_cast_fp16")];
            tensor<string, []> var_1260_equation_0 = const()[name = tensor<string, []>("op_1260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1260_cast_fp16 = einsum(equation = var_1260_equation_0, values = (var_1182_cast_fp16_6, var_1237_cast_fp16))[name = tensor<string, []>("op_1260_cast_fp16")];
            tensor<string, []> var_1262_equation_0 = const()[name = tensor<string, []>("op_1262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1262_cast_fp16 = einsum(equation = var_1262_equation_0, values = (var_1182_cast_fp16_7, var_1238_cast_fp16))[name = tensor<string, []>("op_1262_cast_fp16")];
            tensor<string, []> var_1264_equation_0 = const()[name = tensor<string, []>("op_1264_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1264_cast_fp16 = einsum(equation = var_1264_equation_0, values = (var_1182_cast_fp16_8, var_1239_cast_fp16))[name = tensor<string, []>("op_1264_cast_fp16")];
            tensor<string, []> var_1266_equation_0 = const()[name = tensor<string, []>("op_1266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1266_cast_fp16 = einsum(equation = var_1266_equation_0, values = (var_1182_cast_fp16_9, var_1240_cast_fp16))[name = tensor<string, []>("op_1266_cast_fp16")];
            tensor<string, []> var_1268_equation_0 = const()[name = tensor<string, []>("op_1268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1268_cast_fp16 = einsum(equation = var_1268_equation_0, values = (var_1182_cast_fp16_10, var_1241_cast_fp16))[name = tensor<string, []>("op_1268_cast_fp16")];
            tensor<string, []> var_1270_equation_0 = const()[name = tensor<string, []>("op_1270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1270_cast_fp16 = einsum(equation = var_1270_equation_0, values = (var_1182_cast_fp16_11, var_1242_cast_fp16))[name = tensor<string, []>("op_1270_cast_fp16")];
            tensor<string, []> var_1272_equation_0 = const()[name = tensor<string, []>("op_1272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1272_cast_fp16 = einsum(equation = var_1272_equation_0, values = (var_1182_cast_fp16_12, var_1243_cast_fp16))[name = tensor<string, []>("op_1272_cast_fp16")];
            tensor<string, []> var_1274_equation_0 = const()[name = tensor<string, []>("op_1274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1274_cast_fp16 = einsum(equation = var_1274_equation_0, values = (var_1182_cast_fp16_13, var_1244_cast_fp16))[name = tensor<string, []>("op_1274_cast_fp16")];
            tensor<string, []> var_1276_equation_0 = const()[name = tensor<string, []>("op_1276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1276_cast_fp16 = einsum(equation = var_1276_equation_0, values = (var_1182_cast_fp16_14, var_1245_cast_fp16))[name = tensor<string, []>("op_1276_cast_fp16")];
            tensor<string, []> var_1278_equation_0 = const()[name = tensor<string, []>("op_1278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1278_cast_fp16 = einsum(equation = var_1278_equation_0, values = (var_1182_cast_fp16_15, var_1246_cast_fp16))[name = tensor<string, []>("op_1278_cast_fp16")];
            tensor<bool, []> input_47_interleave_0 = const()[name = tensor<string, []>("input_47_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_47_cast_fp16 = concat(axis = var_1095, interleave = input_47_interleave_0, values = (var_1248_cast_fp16, var_1250_cast_fp16, var_1252_cast_fp16, var_1254_cast_fp16, var_1256_cast_fp16, var_1258_cast_fp16, var_1260_cast_fp16, var_1262_cast_fp16, var_1264_cast_fp16, var_1266_cast_fp16, var_1268_cast_fp16, var_1270_cast_fp16, var_1272_cast_fp16, var_1274_cast_fp16, var_1276_cast_fp16, var_1278_cast_fp16))[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<string, []> var_1287_pad_type_0 = const()[name = tensor<string, []>("op_1287_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1287_strides_0 = const()[name = tensor<string, []>("op_1287_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1287_pad_0 = const()[name = tensor<string, []>("op_1287_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1287_dilations_0 = const()[name = tensor<string, []>("op_1287_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1287_groups_0 = const()[name = tensor<string, []>("op_1287_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_4_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28860736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29385088))), name = tensor<string, []>("blocks_4_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_4_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29385216)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1287_cast_fp16 = conv(bias = blocks_4_attn_out_bias_to_fp16, dilations = var_1287_dilations_0, groups = var_1287_groups_0, pad = var_1287_pad_0, pad_type = var_1287_pad_type_0, strides = var_1287_strides_0, weight = blocks_4_attn_out_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = tensor<string, []>("op_1287_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_1287_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")];
            tensor<int32, [1]> input_49_axes_0 = const()[name = tensor<string, []>("input_49_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_49_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_49_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29387328)))];
            tensor<fp16, [1024]> input_49_beta_0_to_fp16 = const()[name = tensor<string, []>("input_49_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29389440)))];
            tensor<fp16, []> var_1297_to_fp16 = const()[name = tensor<string, []>("op_1297_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_49_cast_fp16 = layer_norm(axes = input_49_axes_0, beta = input_49_beta_0_to_fp16, epsilon = var_1297_to_fp16, gamma = input_49_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<string, []> input_51_pad_type_0 = const()[name = tensor<string, []>("input_51_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_51_strides_0 = const()[name = tensor<string, []>("input_51_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_51_pad_0 = const()[name = tensor<string, []>("input_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_51_dilations_0 = const()[name = tensor<string, []>("input_51_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_51_groups_0 = const()[name = tensor<string, []>("input_51_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_4_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29391552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31488768))), name = tensor<string, []>("blocks_4_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_4_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31488896)))];
            tensor<fp16, [1, 4096, 1, 600]> input_51_cast_fp16 = conv(bias = blocks_4_mlp_0_bias_to_fp16, dilations = input_51_dilations_0, groups = input_51_groups_0, pad = input_51_pad_0, pad_type = input_51_pad_type_0, strides = input_51_strides_0, weight = blocks_4_mlp_0_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
            tensor<string, []> input_53_mode_0 = const()[name = tensor<string, []>("input_53_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_53_cast_fp16 = gelu(mode = input_53_mode_0, x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<string, []> var_1323_pad_type_0 = const()[name = tensor<string, []>("op_1323_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1323_strides_0 = const()[name = tensor<string, []>("op_1323_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1323_pad_0 = const()[name = tensor<string, []>("op_1323_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1323_dilations_0 = const()[name = tensor<string, []>("op_1323_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1323_groups_0 = const()[name = tensor<string, []>("op_1323_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_4_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31497152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33594368))), name = tensor<string, []>("blocks_4_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_4_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33594496)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1323_cast_fp16 = conv(bias = blocks_4_mlp_2_bias_to_fp16, dilations = var_1323_dilations_0, groups = var_1323_groups_0, pad = var_1323_pad_0, pad_type = var_1323_pad_type_0, strides = var_1323_strides_0, weight = blocks_4_mlp_2_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = tensor<string, []>("op_1323_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = var_1323_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")];
            tensor<int32, []> var_1332 = const()[name = tensor<string, []>("op_1332"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_55_axes_0 = const()[name = tensor<string, []>("input_55_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_55_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33596608)))];
            tensor<fp16, [1024]> input_55_beta_0_to_fp16 = const()[name = tensor<string, []>("input_55_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33598720)))];
            tensor<fp16, []> var_1348_to_fp16 = const()[name = tensor<string, []>("op_1348_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_55_cast_fp16 = layer_norm(axes = input_55_axes_0, beta = input_55_beta_0_to_fp16, epsilon = var_1348_to_fp16, gamma = input_55_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<string, []> q_11_pad_type_0 = const()[name = tensor<string, []>("q_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_11_strides_0 = const()[name = tensor<string, []>("q_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_11_pad_0 = const()[name = tensor<string, []>("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_11_dilations_0 = const()[name = tensor<string, []>("q_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_11_groups_0 = const()[name = tensor<string, []>("q_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_1383_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33600832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34125184))), name = tensor<string, []>("op_1383_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_1383_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1383_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34125312)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1383_cast_fp16 = conv(bias = var_1383_bias_0_to_fp16, dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = op_1383_weight_0_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("op_1383_cast_fp16")];
            tensor<string, []> k_11_pad_type_0 = const()[name = tensor<string, []>("k_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_11_strides_0 = const()[name = tensor<string, []>("k_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_11_pad_0 = const()[name = tensor<string, []>("k_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_11_dilations_0 = const()[name = tensor<string, []>("k_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_11_groups_0 = const()[name = tensor<string, []>("k_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_5_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34127424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34651776))), name = tensor<string, []>("blocks_5_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_11_cast_fp16 = conv(dilations = k_11_dilations_0, groups = k_11_groups_0, pad = k_11_pad_0, pad_type = k_11_pad_type_0, strides = k_11_strides_0, weight = blocks_5_attn_key_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("k_11_cast_fp16")];
            tensor<string, []> var_1381_pad_type_0 = const()[name = tensor<string, []>("op_1381_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1381_strides_0 = const()[name = tensor<string, []>("op_1381_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1381_pad_0 = const()[name = tensor<string, []>("op_1381_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1381_dilations_0 = const()[name = tensor<string, []>("op_1381_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1381_groups_0 = const()[name = tensor<string, []>("op_1381_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_5_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34651904))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35176256))), name = tensor<string, []>("blocks_5_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_5_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35176384)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1381_cast_fp16 = conv(bias = blocks_5_attn_value_bias_to_fp16, dilations = var_1381_dilations_0, groups = var_1381_groups_0, pad = var_1381_pad_0, pad_type = var_1381_pad_type_0, strides = var_1381_strides_0, weight = blocks_5_attn_value_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("op_1381_cast_fp16")];
            tensor<int32, [16]> tile_15 = const()[name = tensor<string, []>("tile_15"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1384_axis_0 = const()[name = tensor<string, []>("op_1384_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1384_cast_fp16_15 = split(axis = var_1384_axis_0, split_sizes = tile_15, x = var_1383_cast_fp16)[name = tensor<string, []>("op_1384_cast_fp16")];
            tensor<int32, [4]> var_1401_perm_0 = const()[name = tensor<string, []>("op_1401_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_16 = const()[name = tensor<string, []>("tile_16"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1402_axis_0 = const()[name = tensor<string, []>("op_1402_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_1401_cast_fp16 = transpose(perm = var_1401_perm_0, x = k_11_cast_fp16)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_1402_cast_fp16_15 = split(axis = var_1402_axis_0, split_sizes = tile_16, x = var_1401_cast_fp16)[name = tensor<string, []>("op_1402_cast_fp16")];
            tensor<int32, [16]> tile_17 = const()[name = tensor<string, []>("tile_17"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1419_axis_0 = const()[name = tensor<string, []>("op_1419_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1419_cast_fp16_15 = split(axis = var_1419_axis_0, split_sizes = tile_17, x = var_1381_cast_fp16)[name = tensor<string, []>("op_1419_cast_fp16")];
            tensor<string, []> aw_161_equation_0 = const()[name = tensor<string, []>("aw_161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_161_cast_fp16 = einsum(equation = aw_161_equation_0, values = (var_1402_cast_fp16_0, var_1384_cast_fp16_0))[name = tensor<string, []>("aw_161_cast_fp16")];
            tensor<string, []> aw_163_equation_0 = const()[name = tensor<string, []>("aw_163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_163_cast_fp16 = einsum(equation = aw_163_equation_0, values = (var_1402_cast_fp16_1, var_1384_cast_fp16_1))[name = tensor<string, []>("aw_163_cast_fp16")];
            tensor<string, []> aw_165_equation_0 = const()[name = tensor<string, []>("aw_165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_165_cast_fp16 = einsum(equation = aw_165_equation_0, values = (var_1402_cast_fp16_2, var_1384_cast_fp16_2))[name = tensor<string, []>("aw_165_cast_fp16")];
            tensor<string, []> aw_167_equation_0 = const()[name = tensor<string, []>("aw_167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_167_cast_fp16 = einsum(equation = aw_167_equation_0, values = (var_1402_cast_fp16_3, var_1384_cast_fp16_3))[name = tensor<string, []>("aw_167_cast_fp16")];
            tensor<string, []> aw_169_equation_0 = const()[name = tensor<string, []>("aw_169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_169_cast_fp16 = einsum(equation = aw_169_equation_0, values = (var_1402_cast_fp16_4, var_1384_cast_fp16_4))[name = tensor<string, []>("aw_169_cast_fp16")];
            tensor<string, []> aw_171_equation_0 = const()[name = tensor<string, []>("aw_171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_171_cast_fp16 = einsum(equation = aw_171_equation_0, values = (var_1402_cast_fp16_5, var_1384_cast_fp16_5))[name = tensor<string, []>("aw_171_cast_fp16")];
            tensor<string, []> aw_173_equation_0 = const()[name = tensor<string, []>("aw_173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_173_cast_fp16 = einsum(equation = aw_173_equation_0, values = (var_1402_cast_fp16_6, var_1384_cast_fp16_6))[name = tensor<string, []>("aw_173_cast_fp16")];
            tensor<string, []> aw_175_equation_0 = const()[name = tensor<string, []>("aw_175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_175_cast_fp16 = einsum(equation = aw_175_equation_0, values = (var_1402_cast_fp16_7, var_1384_cast_fp16_7))[name = tensor<string, []>("aw_175_cast_fp16")];
            tensor<string, []> aw_177_equation_0 = const()[name = tensor<string, []>("aw_177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_177_cast_fp16 = einsum(equation = aw_177_equation_0, values = (var_1402_cast_fp16_8, var_1384_cast_fp16_8))[name = tensor<string, []>("aw_177_cast_fp16")];
            tensor<string, []> aw_179_equation_0 = const()[name = tensor<string, []>("aw_179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_179_cast_fp16 = einsum(equation = aw_179_equation_0, values = (var_1402_cast_fp16_9, var_1384_cast_fp16_9))[name = tensor<string, []>("aw_179_cast_fp16")];
            tensor<string, []> aw_181_equation_0 = const()[name = tensor<string, []>("aw_181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_181_cast_fp16 = einsum(equation = aw_181_equation_0, values = (var_1402_cast_fp16_10, var_1384_cast_fp16_10))[name = tensor<string, []>("aw_181_cast_fp16")];
            tensor<string, []> aw_183_equation_0 = const()[name = tensor<string, []>("aw_183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_183_cast_fp16 = einsum(equation = aw_183_equation_0, values = (var_1402_cast_fp16_11, var_1384_cast_fp16_11))[name = tensor<string, []>("aw_183_cast_fp16")];
            tensor<string, []> aw_185_equation_0 = const()[name = tensor<string, []>("aw_185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_185_cast_fp16 = einsum(equation = aw_185_equation_0, values = (var_1402_cast_fp16_12, var_1384_cast_fp16_12))[name = tensor<string, []>("aw_185_cast_fp16")];
            tensor<string, []> aw_187_equation_0 = const()[name = tensor<string, []>("aw_187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_187_cast_fp16 = einsum(equation = aw_187_equation_0, values = (var_1402_cast_fp16_13, var_1384_cast_fp16_13))[name = tensor<string, []>("aw_187_cast_fp16")];
            tensor<string, []> aw_189_equation_0 = const()[name = tensor<string, []>("aw_189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_189_cast_fp16 = einsum(equation = aw_189_equation_0, values = (var_1402_cast_fp16_14, var_1384_cast_fp16_14))[name = tensor<string, []>("aw_189_cast_fp16")];
            tensor<string, []> aw_191_equation_0 = const()[name = tensor<string, []>("aw_191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_191_cast_fp16 = einsum(equation = aw_191_equation_0, values = (var_1402_cast_fp16_15, var_1384_cast_fp16_15))[name = tensor<string, []>("aw_191_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1468_cast_fp16 = softmax(axis = var_1332, x = aw_161_cast_fp16)[name = tensor<string, []>("op_1468_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1469_cast_fp16 = softmax(axis = var_1332, x = aw_163_cast_fp16)[name = tensor<string, []>("op_1469_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1470_cast_fp16 = softmax(axis = var_1332, x = aw_165_cast_fp16)[name = tensor<string, []>("op_1470_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1471_cast_fp16 = softmax(axis = var_1332, x = aw_167_cast_fp16)[name = tensor<string, []>("op_1471_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1472_cast_fp16 = softmax(axis = var_1332, x = aw_169_cast_fp16)[name = tensor<string, []>("op_1472_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1473_cast_fp16 = softmax(axis = var_1332, x = aw_171_cast_fp16)[name = tensor<string, []>("op_1473_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1474_cast_fp16 = softmax(axis = var_1332, x = aw_173_cast_fp16)[name = tensor<string, []>("op_1474_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1475_cast_fp16 = softmax(axis = var_1332, x = aw_175_cast_fp16)[name = tensor<string, []>("op_1475_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1476_cast_fp16 = softmax(axis = var_1332, x = aw_177_cast_fp16)[name = tensor<string, []>("op_1476_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1477_cast_fp16 = softmax(axis = var_1332, x = aw_179_cast_fp16)[name = tensor<string, []>("op_1477_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1478_cast_fp16 = softmax(axis = var_1332, x = aw_181_cast_fp16)[name = tensor<string, []>("op_1478_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1479_cast_fp16 = softmax(axis = var_1332, x = aw_183_cast_fp16)[name = tensor<string, []>("op_1479_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1480_cast_fp16 = softmax(axis = var_1332, x = aw_185_cast_fp16)[name = tensor<string, []>("op_1480_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1481_cast_fp16 = softmax(axis = var_1332, x = aw_187_cast_fp16)[name = tensor<string, []>("op_1481_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1482_cast_fp16 = softmax(axis = var_1332, x = aw_189_cast_fp16)[name = tensor<string, []>("op_1482_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1483_cast_fp16 = softmax(axis = var_1332, x = aw_191_cast_fp16)[name = tensor<string, []>("op_1483_cast_fp16")];
            tensor<string, []> var_1485_equation_0 = const()[name = tensor<string, []>("op_1485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1485_cast_fp16 = einsum(equation = var_1485_equation_0, values = (var_1419_cast_fp16_0, var_1468_cast_fp16))[name = tensor<string, []>("op_1485_cast_fp16")];
            tensor<string, []> var_1487_equation_0 = const()[name = tensor<string, []>("op_1487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1487_cast_fp16 = einsum(equation = var_1487_equation_0, values = (var_1419_cast_fp16_1, var_1469_cast_fp16))[name = tensor<string, []>("op_1487_cast_fp16")];
            tensor<string, []> var_1489_equation_0 = const()[name = tensor<string, []>("op_1489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1489_cast_fp16 = einsum(equation = var_1489_equation_0, values = (var_1419_cast_fp16_2, var_1470_cast_fp16))[name = tensor<string, []>("op_1489_cast_fp16")];
            tensor<string, []> var_1491_equation_0 = const()[name = tensor<string, []>("op_1491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1491_cast_fp16 = einsum(equation = var_1491_equation_0, values = (var_1419_cast_fp16_3, var_1471_cast_fp16))[name = tensor<string, []>("op_1491_cast_fp16")];
            tensor<string, []> var_1493_equation_0 = const()[name = tensor<string, []>("op_1493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1493_cast_fp16 = einsum(equation = var_1493_equation_0, values = (var_1419_cast_fp16_4, var_1472_cast_fp16))[name = tensor<string, []>("op_1493_cast_fp16")];
            tensor<string, []> var_1495_equation_0 = const()[name = tensor<string, []>("op_1495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1495_cast_fp16 = einsum(equation = var_1495_equation_0, values = (var_1419_cast_fp16_5, var_1473_cast_fp16))[name = tensor<string, []>("op_1495_cast_fp16")];
            tensor<string, []> var_1497_equation_0 = const()[name = tensor<string, []>("op_1497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1497_cast_fp16 = einsum(equation = var_1497_equation_0, values = (var_1419_cast_fp16_6, var_1474_cast_fp16))[name = tensor<string, []>("op_1497_cast_fp16")];
            tensor<string, []> var_1499_equation_0 = const()[name = tensor<string, []>("op_1499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1499_cast_fp16 = einsum(equation = var_1499_equation_0, values = (var_1419_cast_fp16_7, var_1475_cast_fp16))[name = tensor<string, []>("op_1499_cast_fp16")];
            tensor<string, []> var_1501_equation_0 = const()[name = tensor<string, []>("op_1501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1501_cast_fp16 = einsum(equation = var_1501_equation_0, values = (var_1419_cast_fp16_8, var_1476_cast_fp16))[name = tensor<string, []>("op_1501_cast_fp16")];
            tensor<string, []> var_1503_equation_0 = const()[name = tensor<string, []>("op_1503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1503_cast_fp16 = einsum(equation = var_1503_equation_0, values = (var_1419_cast_fp16_9, var_1477_cast_fp16))[name = tensor<string, []>("op_1503_cast_fp16")];
            tensor<string, []> var_1505_equation_0 = const()[name = tensor<string, []>("op_1505_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1505_cast_fp16 = einsum(equation = var_1505_equation_0, values = (var_1419_cast_fp16_10, var_1478_cast_fp16))[name = tensor<string, []>("op_1505_cast_fp16")];
            tensor<string, []> var_1507_equation_0 = const()[name = tensor<string, []>("op_1507_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1507_cast_fp16 = einsum(equation = var_1507_equation_0, values = (var_1419_cast_fp16_11, var_1479_cast_fp16))[name = tensor<string, []>("op_1507_cast_fp16")];
            tensor<string, []> var_1509_equation_0 = const()[name = tensor<string, []>("op_1509_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1419_cast_fp16_12, var_1480_cast_fp16))[name = tensor<string, []>("op_1509_cast_fp16")];
            tensor<string, []> var_1511_equation_0 = const()[name = tensor<string, []>("op_1511_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1419_cast_fp16_13, var_1481_cast_fp16))[name = tensor<string, []>("op_1511_cast_fp16")];
            tensor<string, []> var_1513_equation_0 = const()[name = tensor<string, []>("op_1513_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1419_cast_fp16_14, var_1482_cast_fp16))[name = tensor<string, []>("op_1513_cast_fp16")];
            tensor<string, []> var_1515_equation_0 = const()[name = tensor<string, []>("op_1515_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1419_cast_fp16_15, var_1483_cast_fp16))[name = tensor<string, []>("op_1515_cast_fp16")];
            tensor<bool, []> input_57_interleave_0 = const()[name = tensor<string, []>("input_57_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_57_cast_fp16 = concat(axis = var_1332, interleave = input_57_interleave_0, values = (var_1485_cast_fp16, var_1487_cast_fp16, var_1489_cast_fp16, var_1491_cast_fp16, var_1493_cast_fp16, var_1495_cast_fp16, var_1497_cast_fp16, var_1499_cast_fp16, var_1501_cast_fp16, var_1503_cast_fp16, var_1505_cast_fp16, var_1507_cast_fp16, var_1509_cast_fp16, var_1511_cast_fp16, var_1513_cast_fp16, var_1515_cast_fp16))[name = tensor<string, []>("input_57_cast_fp16")];
            tensor<string, []> var_1524_pad_type_0 = const()[name = tensor<string, []>("op_1524_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1524_strides_0 = const()[name = tensor<string, []>("op_1524_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1524_pad_0 = const()[name = tensor<string, []>("op_1524_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1524_dilations_0 = const()[name = tensor<string, []>("op_1524_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1524_groups_0 = const()[name = tensor<string, []>("op_1524_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_5_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35178496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35702848))), name = tensor<string, []>("blocks_5_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_5_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35702976)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1524_cast_fp16 = conv(bias = blocks_5_attn_out_bias_to_fp16, dilations = var_1524_dilations_0, groups = var_1524_groups_0, pad = var_1524_pad_0, pad_type = var_1524_pad_type_0, strides = var_1524_strides_0, weight = blocks_5_attn_out_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor<string, []>("op_1524_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_1524_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")];
            tensor<int32, [1]> input_59_axes_0 = const()[name = tensor<string, []>("input_59_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_59_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35705088)))];
            tensor<fp16, [1024]> input_59_beta_0_to_fp16 = const()[name = tensor<string, []>("input_59_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35707200)))];
            tensor<fp16, []> var_1534_to_fp16 = const()[name = tensor<string, []>("op_1534_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = input_59_beta_0_to_fp16, epsilon = var_1534_to_fp16, gamma = input_59_gamma_0_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<string, []> input_61_pad_type_0 = const()[name = tensor<string, []>("input_61_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_61_strides_0 = const()[name = tensor<string, []>("input_61_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_61_pad_0 = const()[name = tensor<string, []>("input_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_61_dilations_0 = const()[name = tensor<string, []>("input_61_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_61_groups_0 = const()[name = tensor<string, []>("input_61_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_5_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35709312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37806528))), name = tensor<string, []>("blocks_5_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_5_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37806656)))];
            tensor<fp16, [1, 4096, 1, 600]> input_61_cast_fp16 = conv(bias = blocks_5_mlp_0_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = blocks_5_mlp_0_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<string, []> input_63_mode_0 = const()[name = tensor<string, []>("input_63_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
            tensor<string, []> var_1560_pad_type_0 = const()[name = tensor<string, []>("op_1560_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1560_strides_0 = const()[name = tensor<string, []>("op_1560_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1560_pad_0 = const()[name = tensor<string, []>("op_1560_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1560_dilations_0 = const()[name = tensor<string, []>("op_1560_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1560_groups_0 = const()[name = tensor<string, []>("op_1560_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_5_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37814912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39912128))), name = tensor<string, []>("blocks_5_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_5_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39912256)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1560_cast_fp16 = conv(bias = blocks_5_mlp_2_bias_to_fp16, dilations = var_1560_dilations_0, groups = var_1560_groups_0, pad = var_1560_pad_0, pad_type = var_1560_pad_type_0, strides = var_1560_strides_0, weight = blocks_5_mlp_2_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor<string, []>("op_1560_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = var_1560_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")];
            tensor<int32, []> var_1569 = const()[name = tensor<string, []>("op_1569"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_65_axes_0 = const()[name = tensor<string, []>("input_65_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_65_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39914368)))];
            tensor<fp16, [1024]> input_65_beta_0_to_fp16 = const()[name = tensor<string, []>("input_65_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39916480)))];
            tensor<fp16, []> var_1585_to_fp16 = const()[name = tensor<string, []>("op_1585_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_65_cast_fp16 = layer_norm(axes = input_65_axes_0, beta = input_65_beta_0_to_fp16, epsilon = var_1585_to_fp16, gamma = input_65_gamma_0_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")];
            tensor<string, []> q_13_pad_type_0 = const()[name = tensor<string, []>("q_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_13_strides_0 = const()[name = tensor<string, []>("q_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_13_pad_0 = const()[name = tensor<string, []>("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_13_dilations_0 = const()[name = tensor<string, []>("q_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_13_groups_0 = const()[name = tensor<string, []>("q_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_1620_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39918592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40442944))), name = tensor<string, []>("op_1620_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_1620_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1620_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40443072)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1620_cast_fp16 = conv(bias = var_1620_bias_0_to_fp16, dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = op_1620_weight_0_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("op_1620_cast_fp16")];
            tensor<string, []> k_13_pad_type_0 = const()[name = tensor<string, []>("k_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_13_strides_0 = const()[name = tensor<string, []>("k_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_13_pad_0 = const()[name = tensor<string, []>("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_13_dilations_0 = const()[name = tensor<string, []>("k_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_13_groups_0 = const()[name = tensor<string, []>("k_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_6_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40445184))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40969536))), name = tensor<string, []>("blocks_6_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = blocks_6_attn_key_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")];
            tensor<string, []> var_1618_pad_type_0 = const()[name = tensor<string, []>("op_1618_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1618_strides_0 = const()[name = tensor<string, []>("op_1618_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1618_pad_0 = const()[name = tensor<string, []>("op_1618_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1618_dilations_0 = const()[name = tensor<string, []>("op_1618_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1618_groups_0 = const()[name = tensor<string, []>("op_1618_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_6_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40969664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41494016))), name = tensor<string, []>("blocks_6_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_6_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41494144)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1618_cast_fp16 = conv(bias = blocks_6_attn_value_bias_to_fp16, dilations = var_1618_dilations_0, groups = var_1618_groups_0, pad = var_1618_pad_0, pad_type = var_1618_pad_type_0, strides = var_1618_strides_0, weight = blocks_6_attn_value_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("op_1618_cast_fp16")];
            tensor<int32, [16]> tile_18 = const()[name = tensor<string, []>("tile_18"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1621_axis_0 = const()[name = tensor<string, []>("op_1621_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1621_cast_fp16_15 = split(axis = var_1621_axis_0, split_sizes = tile_18, x = var_1620_cast_fp16)[name = tensor<string, []>("op_1621_cast_fp16")];
            tensor<int32, [4]> var_1638_perm_0 = const()[name = tensor<string, []>("op_1638_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_19 = const()[name = tensor<string, []>("tile_19"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1639_axis_0 = const()[name = tensor<string, []>("op_1639_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_1638_cast_fp16 = transpose(perm = var_1638_perm_0, x = k_13_cast_fp16)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_1639_cast_fp16_15 = split(axis = var_1639_axis_0, split_sizes = tile_19, x = var_1638_cast_fp16)[name = tensor<string, []>("op_1639_cast_fp16")];
            tensor<int32, [16]> tile_20 = const()[name = tensor<string, []>("tile_20"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1656_axis_0 = const()[name = tensor<string, []>("op_1656_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1656_cast_fp16_15 = split(axis = var_1656_axis_0, split_sizes = tile_20, x = var_1618_cast_fp16)[name = tensor<string, []>("op_1656_cast_fp16")];
            tensor<string, []> aw_193_equation_0 = const()[name = tensor<string, []>("aw_193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_193_cast_fp16 = einsum(equation = aw_193_equation_0, values = (var_1639_cast_fp16_0, var_1621_cast_fp16_0))[name = tensor<string, []>("aw_193_cast_fp16")];
            tensor<string, []> aw_195_equation_0 = const()[name = tensor<string, []>("aw_195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_195_cast_fp16 = einsum(equation = aw_195_equation_0, values = (var_1639_cast_fp16_1, var_1621_cast_fp16_1))[name = tensor<string, []>("aw_195_cast_fp16")];
            tensor<string, []> aw_197_equation_0 = const()[name = tensor<string, []>("aw_197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_197_cast_fp16 = einsum(equation = aw_197_equation_0, values = (var_1639_cast_fp16_2, var_1621_cast_fp16_2))[name = tensor<string, []>("aw_197_cast_fp16")];
            tensor<string, []> aw_199_equation_0 = const()[name = tensor<string, []>("aw_199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_199_cast_fp16 = einsum(equation = aw_199_equation_0, values = (var_1639_cast_fp16_3, var_1621_cast_fp16_3))[name = tensor<string, []>("aw_199_cast_fp16")];
            tensor<string, []> aw_201_equation_0 = const()[name = tensor<string, []>("aw_201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_201_cast_fp16 = einsum(equation = aw_201_equation_0, values = (var_1639_cast_fp16_4, var_1621_cast_fp16_4))[name = tensor<string, []>("aw_201_cast_fp16")];
            tensor<string, []> aw_203_equation_0 = const()[name = tensor<string, []>("aw_203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_203_cast_fp16 = einsum(equation = aw_203_equation_0, values = (var_1639_cast_fp16_5, var_1621_cast_fp16_5))[name = tensor<string, []>("aw_203_cast_fp16")];
            tensor<string, []> aw_205_equation_0 = const()[name = tensor<string, []>("aw_205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_205_cast_fp16 = einsum(equation = aw_205_equation_0, values = (var_1639_cast_fp16_6, var_1621_cast_fp16_6))[name = tensor<string, []>("aw_205_cast_fp16")];
            tensor<string, []> aw_207_equation_0 = const()[name = tensor<string, []>("aw_207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_207_cast_fp16 = einsum(equation = aw_207_equation_0, values = (var_1639_cast_fp16_7, var_1621_cast_fp16_7))[name = tensor<string, []>("aw_207_cast_fp16")];
            tensor<string, []> aw_209_equation_0 = const()[name = tensor<string, []>("aw_209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_209_cast_fp16 = einsum(equation = aw_209_equation_0, values = (var_1639_cast_fp16_8, var_1621_cast_fp16_8))[name = tensor<string, []>("aw_209_cast_fp16")];
            tensor<string, []> aw_211_equation_0 = const()[name = tensor<string, []>("aw_211_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_211_cast_fp16 = einsum(equation = aw_211_equation_0, values = (var_1639_cast_fp16_9, var_1621_cast_fp16_9))[name = tensor<string, []>("aw_211_cast_fp16")];
            tensor<string, []> aw_213_equation_0 = const()[name = tensor<string, []>("aw_213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_213_cast_fp16 = einsum(equation = aw_213_equation_0, values = (var_1639_cast_fp16_10, var_1621_cast_fp16_10))[name = tensor<string, []>("aw_213_cast_fp16")];
            tensor<string, []> aw_215_equation_0 = const()[name = tensor<string, []>("aw_215_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_215_cast_fp16 = einsum(equation = aw_215_equation_0, values = (var_1639_cast_fp16_11, var_1621_cast_fp16_11))[name = tensor<string, []>("aw_215_cast_fp16")];
            tensor<string, []> aw_217_equation_0 = const()[name = tensor<string, []>("aw_217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_217_cast_fp16 = einsum(equation = aw_217_equation_0, values = (var_1639_cast_fp16_12, var_1621_cast_fp16_12))[name = tensor<string, []>("aw_217_cast_fp16")];
            tensor<string, []> aw_219_equation_0 = const()[name = tensor<string, []>("aw_219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_219_cast_fp16 = einsum(equation = aw_219_equation_0, values = (var_1639_cast_fp16_13, var_1621_cast_fp16_13))[name = tensor<string, []>("aw_219_cast_fp16")];
            tensor<string, []> aw_221_equation_0 = const()[name = tensor<string, []>("aw_221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_221_cast_fp16 = einsum(equation = aw_221_equation_0, values = (var_1639_cast_fp16_14, var_1621_cast_fp16_14))[name = tensor<string, []>("aw_221_cast_fp16")];
            tensor<string, []> aw_223_equation_0 = const()[name = tensor<string, []>("aw_223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_223_cast_fp16 = einsum(equation = aw_223_equation_0, values = (var_1639_cast_fp16_15, var_1621_cast_fp16_15))[name = tensor<string, []>("aw_223_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1705_cast_fp16 = softmax(axis = var_1569, x = aw_193_cast_fp16)[name = tensor<string, []>("op_1705_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1706_cast_fp16 = softmax(axis = var_1569, x = aw_195_cast_fp16)[name = tensor<string, []>("op_1706_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1707_cast_fp16 = softmax(axis = var_1569, x = aw_197_cast_fp16)[name = tensor<string, []>("op_1707_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1708_cast_fp16 = softmax(axis = var_1569, x = aw_199_cast_fp16)[name = tensor<string, []>("op_1708_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1709_cast_fp16 = softmax(axis = var_1569, x = aw_201_cast_fp16)[name = tensor<string, []>("op_1709_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1710_cast_fp16 = softmax(axis = var_1569, x = aw_203_cast_fp16)[name = tensor<string, []>("op_1710_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1711_cast_fp16 = softmax(axis = var_1569, x = aw_205_cast_fp16)[name = tensor<string, []>("op_1711_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1712_cast_fp16 = softmax(axis = var_1569, x = aw_207_cast_fp16)[name = tensor<string, []>("op_1712_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1713_cast_fp16 = softmax(axis = var_1569, x = aw_209_cast_fp16)[name = tensor<string, []>("op_1713_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1714_cast_fp16 = softmax(axis = var_1569, x = aw_211_cast_fp16)[name = tensor<string, []>("op_1714_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1715_cast_fp16 = softmax(axis = var_1569, x = aw_213_cast_fp16)[name = tensor<string, []>("op_1715_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1716_cast_fp16 = softmax(axis = var_1569, x = aw_215_cast_fp16)[name = tensor<string, []>("op_1716_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1717_cast_fp16 = softmax(axis = var_1569, x = aw_217_cast_fp16)[name = tensor<string, []>("op_1717_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1718_cast_fp16 = softmax(axis = var_1569, x = aw_219_cast_fp16)[name = tensor<string, []>("op_1718_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1719_cast_fp16 = softmax(axis = var_1569, x = aw_221_cast_fp16)[name = tensor<string, []>("op_1719_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1720_cast_fp16 = softmax(axis = var_1569, x = aw_223_cast_fp16)[name = tensor<string, []>("op_1720_cast_fp16")];
            tensor<string, []> var_1722_equation_0 = const()[name = tensor<string, []>("op_1722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1722_cast_fp16 = einsum(equation = var_1722_equation_0, values = (var_1656_cast_fp16_0, var_1705_cast_fp16))[name = tensor<string, []>("op_1722_cast_fp16")];
            tensor<string, []> var_1724_equation_0 = const()[name = tensor<string, []>("op_1724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1724_cast_fp16 = einsum(equation = var_1724_equation_0, values = (var_1656_cast_fp16_1, var_1706_cast_fp16))[name = tensor<string, []>("op_1724_cast_fp16")];
            tensor<string, []> var_1726_equation_0 = const()[name = tensor<string, []>("op_1726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1726_cast_fp16 = einsum(equation = var_1726_equation_0, values = (var_1656_cast_fp16_2, var_1707_cast_fp16))[name = tensor<string, []>("op_1726_cast_fp16")];
            tensor<string, []> var_1728_equation_0 = const()[name = tensor<string, []>("op_1728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1728_cast_fp16 = einsum(equation = var_1728_equation_0, values = (var_1656_cast_fp16_3, var_1708_cast_fp16))[name = tensor<string, []>("op_1728_cast_fp16")];
            tensor<string, []> var_1730_equation_0 = const()[name = tensor<string, []>("op_1730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1730_cast_fp16 = einsum(equation = var_1730_equation_0, values = (var_1656_cast_fp16_4, var_1709_cast_fp16))[name = tensor<string, []>("op_1730_cast_fp16")];
            tensor<string, []> var_1732_equation_0 = const()[name = tensor<string, []>("op_1732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1732_cast_fp16 = einsum(equation = var_1732_equation_0, values = (var_1656_cast_fp16_5, var_1710_cast_fp16))[name = tensor<string, []>("op_1732_cast_fp16")];
            tensor<string, []> var_1734_equation_0 = const()[name = tensor<string, []>("op_1734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1734_cast_fp16 = einsum(equation = var_1734_equation_0, values = (var_1656_cast_fp16_6, var_1711_cast_fp16))[name = tensor<string, []>("op_1734_cast_fp16")];
            tensor<string, []> var_1736_equation_0 = const()[name = tensor<string, []>("op_1736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1736_cast_fp16 = einsum(equation = var_1736_equation_0, values = (var_1656_cast_fp16_7, var_1712_cast_fp16))[name = tensor<string, []>("op_1736_cast_fp16")];
            tensor<string, []> var_1738_equation_0 = const()[name = tensor<string, []>("op_1738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1738_cast_fp16 = einsum(equation = var_1738_equation_0, values = (var_1656_cast_fp16_8, var_1713_cast_fp16))[name = tensor<string, []>("op_1738_cast_fp16")];
            tensor<string, []> var_1740_equation_0 = const()[name = tensor<string, []>("op_1740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1740_cast_fp16 = einsum(equation = var_1740_equation_0, values = (var_1656_cast_fp16_9, var_1714_cast_fp16))[name = tensor<string, []>("op_1740_cast_fp16")];
            tensor<string, []> var_1742_equation_0 = const()[name = tensor<string, []>("op_1742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1742_cast_fp16 = einsum(equation = var_1742_equation_0, values = (var_1656_cast_fp16_10, var_1715_cast_fp16))[name = tensor<string, []>("op_1742_cast_fp16")];
            tensor<string, []> var_1744_equation_0 = const()[name = tensor<string, []>("op_1744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1744_cast_fp16 = einsum(equation = var_1744_equation_0, values = (var_1656_cast_fp16_11, var_1716_cast_fp16))[name = tensor<string, []>("op_1744_cast_fp16")];
            tensor<string, []> var_1746_equation_0 = const()[name = tensor<string, []>("op_1746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1746_cast_fp16 = einsum(equation = var_1746_equation_0, values = (var_1656_cast_fp16_12, var_1717_cast_fp16))[name = tensor<string, []>("op_1746_cast_fp16")];
            tensor<string, []> var_1748_equation_0 = const()[name = tensor<string, []>("op_1748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1748_cast_fp16 = einsum(equation = var_1748_equation_0, values = (var_1656_cast_fp16_13, var_1718_cast_fp16))[name = tensor<string, []>("op_1748_cast_fp16")];
            tensor<string, []> var_1750_equation_0 = const()[name = tensor<string, []>("op_1750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1750_cast_fp16 = einsum(equation = var_1750_equation_0, values = (var_1656_cast_fp16_14, var_1719_cast_fp16))[name = tensor<string, []>("op_1750_cast_fp16")];
            tensor<string, []> var_1752_equation_0 = const()[name = tensor<string, []>("op_1752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1752_cast_fp16 = einsum(equation = var_1752_equation_0, values = (var_1656_cast_fp16_15, var_1720_cast_fp16))[name = tensor<string, []>("op_1752_cast_fp16")];
            tensor<bool, []> input_67_interleave_0 = const()[name = tensor<string, []>("input_67_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_67_cast_fp16 = concat(axis = var_1569, interleave = input_67_interleave_0, values = (var_1722_cast_fp16, var_1724_cast_fp16, var_1726_cast_fp16, var_1728_cast_fp16, var_1730_cast_fp16, var_1732_cast_fp16, var_1734_cast_fp16, var_1736_cast_fp16, var_1738_cast_fp16, var_1740_cast_fp16, var_1742_cast_fp16, var_1744_cast_fp16, var_1746_cast_fp16, var_1748_cast_fp16, var_1750_cast_fp16, var_1752_cast_fp16))[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<string, []> var_1761_pad_type_0 = const()[name = tensor<string, []>("op_1761_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1761_strides_0 = const()[name = tensor<string, []>("op_1761_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1761_pad_0 = const()[name = tensor<string, []>("op_1761_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1761_dilations_0 = const()[name = tensor<string, []>("op_1761_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1761_groups_0 = const()[name = tensor<string, []>("op_1761_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_6_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41496256))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42020608))), name = tensor<string, []>("blocks_6_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_6_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42020736)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1761_cast_fp16 = conv(bias = blocks_6_attn_out_bias_to_fp16, dilations = var_1761_dilations_0, groups = var_1761_groups_0, pad = var_1761_pad_0, pad_type = var_1761_pad_type_0, strides = var_1761_strides_0, weight = blocks_6_attn_out_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor<string, []>("op_1761_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = var_1761_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")];
            tensor<int32, [1]> input_69_axes_0 = const()[name = tensor<string, []>("input_69_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_69_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_69_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42022848)))];
            tensor<fp16, [1024]> input_69_beta_0_to_fp16 = const()[name = tensor<string, []>("input_69_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42024960)))];
            tensor<fp16, []> var_1771_to_fp16 = const()[name = tensor<string, []>("op_1771_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_69_cast_fp16 = layer_norm(axes = input_69_axes_0, beta = input_69_beta_0_to_fp16, epsilon = var_1771_to_fp16, gamma = input_69_gamma_0_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<string, []> input_71_pad_type_0 = const()[name = tensor<string, []>("input_71_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_71_strides_0 = const()[name = tensor<string, []>("input_71_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_71_pad_0 = const()[name = tensor<string, []>("input_71_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_71_dilations_0 = const()[name = tensor<string, []>("input_71_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_71_groups_0 = const()[name = tensor<string, []>("input_71_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_6_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42027072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44124288))), name = tensor<string, []>("blocks_6_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_6_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44124416)))];
            tensor<fp16, [1, 4096, 1, 600]> input_71_cast_fp16 = conv(bias = blocks_6_mlp_0_bias_to_fp16, dilations = input_71_dilations_0, groups = input_71_groups_0, pad = input_71_pad_0, pad_type = input_71_pad_type_0, strides = input_71_strides_0, weight = blocks_6_mlp_0_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
            tensor<string, []> input_73_mode_0 = const()[name = tensor<string, []>("input_73_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_73_cast_fp16 = gelu(mode = input_73_mode_0, x = input_71_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<string, []> var_1797_pad_type_0 = const()[name = tensor<string, []>("op_1797_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1797_strides_0 = const()[name = tensor<string, []>("op_1797_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1797_pad_0 = const()[name = tensor<string, []>("op_1797_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1797_dilations_0 = const()[name = tensor<string, []>("op_1797_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1797_groups_0 = const()[name = tensor<string, []>("op_1797_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_6_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44132672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46229888))), name = tensor<string, []>("blocks_6_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_6_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46230016)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1797_cast_fp16 = conv(bias = blocks_6_mlp_2_bias_to_fp16, dilations = var_1797_dilations_0, groups = var_1797_groups_0, pad = var_1797_pad_0, pad_type = var_1797_pad_type_0, strides = var_1797_strides_0, weight = blocks_6_mlp_2_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("op_1797_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1797_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")];
            tensor<int32, []> var_1806 = const()[name = tensor<string, []>("op_1806"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_75_axes_0 = const()[name = tensor<string, []>("input_75_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_75_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46232128)))];
            tensor<fp16, [1024]> input_75_beta_0_to_fp16 = const()[name = tensor<string, []>("input_75_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46234240)))];
            tensor<fp16, []> var_1822_to_fp16 = const()[name = tensor<string, []>("op_1822_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_75_cast_fp16 = layer_norm(axes = input_75_axes_0, beta = input_75_beta_0_to_fp16, epsilon = var_1822_to_fp16, gamma = input_75_gamma_0_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<string, []> q_15_pad_type_0 = const()[name = tensor<string, []>("q_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_15_strides_0 = const()[name = tensor<string, []>("q_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_15_pad_0 = const()[name = tensor<string, []>("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_15_dilations_0 = const()[name = tensor<string, []>("q_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_15_groups_0 = const()[name = tensor<string, []>("q_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_1857_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46236352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46760704))), name = tensor<string, []>("op_1857_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_1857_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1857_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46760832)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1857_cast_fp16 = conv(bias = var_1857_bias_0_to_fp16, dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = op_1857_weight_0_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor<string, []>("op_1857_cast_fp16")];
            tensor<string, []> k_15_pad_type_0 = const()[name = tensor<string, []>("k_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_15_strides_0 = const()[name = tensor<string, []>("k_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_15_pad_0 = const()[name = tensor<string, []>("k_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_15_dilations_0 = const()[name = tensor<string, []>("k_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_15_groups_0 = const()[name = tensor<string, []>("k_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_7_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46762944))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47287296))), name = tensor<string, []>("blocks_7_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_15_cast_fp16 = conv(dilations = k_15_dilations_0, groups = k_15_groups_0, pad = k_15_pad_0, pad_type = k_15_pad_type_0, strides = k_15_strides_0, weight = blocks_7_attn_key_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor<string, []>("k_15_cast_fp16")];
            tensor<string, []> var_1855_pad_type_0 = const()[name = tensor<string, []>("op_1855_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1855_strides_0 = const()[name = tensor<string, []>("op_1855_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1855_pad_0 = const()[name = tensor<string, []>("op_1855_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1855_dilations_0 = const()[name = tensor<string, []>("op_1855_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1855_groups_0 = const()[name = tensor<string, []>("op_1855_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_7_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47287424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47811776))), name = tensor<string, []>("blocks_7_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_7_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47811904)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1855_cast_fp16 = conv(bias = blocks_7_attn_value_bias_to_fp16, dilations = var_1855_dilations_0, groups = var_1855_groups_0, pad = var_1855_pad_0, pad_type = var_1855_pad_type_0, strides = var_1855_strides_0, weight = blocks_7_attn_value_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor<string, []>("op_1855_cast_fp16")];
            tensor<int32, [16]> tile_21 = const()[name = tensor<string, []>("tile_21"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1858_axis_0 = const()[name = tensor<string, []>("op_1858_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1858_cast_fp16_15 = split(axis = var_1858_axis_0, split_sizes = tile_21, x = var_1857_cast_fp16)[name = tensor<string, []>("op_1858_cast_fp16")];
            tensor<int32, [4]> var_1875_perm_0 = const()[name = tensor<string, []>("op_1875_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_22 = const()[name = tensor<string, []>("tile_22"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1876_axis_0 = const()[name = tensor<string, []>("op_1876_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_1875_cast_fp16 = transpose(perm = var_1875_perm_0, x = k_15_cast_fp16)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_1876_cast_fp16_15 = split(axis = var_1876_axis_0, split_sizes = tile_22, x = var_1875_cast_fp16)[name = tensor<string, []>("op_1876_cast_fp16")];
            tensor<int32, [16]> tile_23 = const()[name = tensor<string, []>("tile_23"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_1893_axis_0 = const()[name = tensor<string, []>("op_1893_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_1893_cast_fp16_15 = split(axis = var_1893_axis_0, split_sizes = tile_23, x = var_1855_cast_fp16)[name = tensor<string, []>("op_1893_cast_fp16")];
            tensor<string, []> aw_225_equation_0 = const()[name = tensor<string, []>("aw_225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_225_cast_fp16 = einsum(equation = aw_225_equation_0, values = (var_1876_cast_fp16_0, var_1858_cast_fp16_0))[name = tensor<string, []>("aw_225_cast_fp16")];
            tensor<string, []> aw_227_equation_0 = const()[name = tensor<string, []>("aw_227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_227_cast_fp16 = einsum(equation = aw_227_equation_0, values = (var_1876_cast_fp16_1, var_1858_cast_fp16_1))[name = tensor<string, []>("aw_227_cast_fp16")];
            tensor<string, []> aw_229_equation_0 = const()[name = tensor<string, []>("aw_229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_229_cast_fp16 = einsum(equation = aw_229_equation_0, values = (var_1876_cast_fp16_2, var_1858_cast_fp16_2))[name = tensor<string, []>("aw_229_cast_fp16")];
            tensor<string, []> aw_231_equation_0 = const()[name = tensor<string, []>("aw_231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_231_cast_fp16 = einsum(equation = aw_231_equation_0, values = (var_1876_cast_fp16_3, var_1858_cast_fp16_3))[name = tensor<string, []>("aw_231_cast_fp16")];
            tensor<string, []> aw_233_equation_0 = const()[name = tensor<string, []>("aw_233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_233_cast_fp16 = einsum(equation = aw_233_equation_0, values = (var_1876_cast_fp16_4, var_1858_cast_fp16_4))[name = tensor<string, []>("aw_233_cast_fp16")];
            tensor<string, []> aw_235_equation_0 = const()[name = tensor<string, []>("aw_235_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_235_cast_fp16 = einsum(equation = aw_235_equation_0, values = (var_1876_cast_fp16_5, var_1858_cast_fp16_5))[name = tensor<string, []>("aw_235_cast_fp16")];
            tensor<string, []> aw_237_equation_0 = const()[name = tensor<string, []>("aw_237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_237_cast_fp16 = einsum(equation = aw_237_equation_0, values = (var_1876_cast_fp16_6, var_1858_cast_fp16_6))[name = tensor<string, []>("aw_237_cast_fp16")];
            tensor<string, []> aw_239_equation_0 = const()[name = tensor<string, []>("aw_239_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_239_cast_fp16 = einsum(equation = aw_239_equation_0, values = (var_1876_cast_fp16_7, var_1858_cast_fp16_7))[name = tensor<string, []>("aw_239_cast_fp16")];
            tensor<string, []> aw_241_equation_0 = const()[name = tensor<string, []>("aw_241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_241_cast_fp16 = einsum(equation = aw_241_equation_0, values = (var_1876_cast_fp16_8, var_1858_cast_fp16_8))[name = tensor<string, []>("aw_241_cast_fp16")];
            tensor<string, []> aw_243_equation_0 = const()[name = tensor<string, []>("aw_243_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_243_cast_fp16 = einsum(equation = aw_243_equation_0, values = (var_1876_cast_fp16_9, var_1858_cast_fp16_9))[name = tensor<string, []>("aw_243_cast_fp16")];
            tensor<string, []> aw_245_equation_0 = const()[name = tensor<string, []>("aw_245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_245_cast_fp16 = einsum(equation = aw_245_equation_0, values = (var_1876_cast_fp16_10, var_1858_cast_fp16_10))[name = tensor<string, []>("aw_245_cast_fp16")];
            tensor<string, []> aw_247_equation_0 = const()[name = tensor<string, []>("aw_247_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_247_cast_fp16 = einsum(equation = aw_247_equation_0, values = (var_1876_cast_fp16_11, var_1858_cast_fp16_11))[name = tensor<string, []>("aw_247_cast_fp16")];
            tensor<string, []> aw_249_equation_0 = const()[name = tensor<string, []>("aw_249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_249_cast_fp16 = einsum(equation = aw_249_equation_0, values = (var_1876_cast_fp16_12, var_1858_cast_fp16_12))[name = tensor<string, []>("aw_249_cast_fp16")];
            tensor<string, []> aw_251_equation_0 = const()[name = tensor<string, []>("aw_251_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_251_cast_fp16 = einsum(equation = aw_251_equation_0, values = (var_1876_cast_fp16_13, var_1858_cast_fp16_13))[name = tensor<string, []>("aw_251_cast_fp16")];
            tensor<string, []> aw_253_equation_0 = const()[name = tensor<string, []>("aw_253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_253_cast_fp16 = einsum(equation = aw_253_equation_0, values = (var_1876_cast_fp16_14, var_1858_cast_fp16_14))[name = tensor<string, []>("aw_253_cast_fp16")];
            tensor<string, []> aw_255_equation_0 = const()[name = tensor<string, []>("aw_255_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_255_cast_fp16 = einsum(equation = aw_255_equation_0, values = (var_1876_cast_fp16_15, var_1858_cast_fp16_15))[name = tensor<string, []>("aw_255_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1942_cast_fp16 = softmax(axis = var_1806, x = aw_225_cast_fp16)[name = tensor<string, []>("op_1942_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1943_cast_fp16 = softmax(axis = var_1806, x = aw_227_cast_fp16)[name = tensor<string, []>("op_1943_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1944_cast_fp16 = softmax(axis = var_1806, x = aw_229_cast_fp16)[name = tensor<string, []>("op_1944_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1945_cast_fp16 = softmax(axis = var_1806, x = aw_231_cast_fp16)[name = tensor<string, []>("op_1945_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1946_cast_fp16 = softmax(axis = var_1806, x = aw_233_cast_fp16)[name = tensor<string, []>("op_1946_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1947_cast_fp16 = softmax(axis = var_1806, x = aw_235_cast_fp16)[name = tensor<string, []>("op_1947_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1948_cast_fp16 = softmax(axis = var_1806, x = aw_237_cast_fp16)[name = tensor<string, []>("op_1948_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1949_cast_fp16 = softmax(axis = var_1806, x = aw_239_cast_fp16)[name = tensor<string, []>("op_1949_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1950_cast_fp16 = softmax(axis = var_1806, x = aw_241_cast_fp16)[name = tensor<string, []>("op_1950_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1951_cast_fp16 = softmax(axis = var_1806, x = aw_243_cast_fp16)[name = tensor<string, []>("op_1951_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1952_cast_fp16 = softmax(axis = var_1806, x = aw_245_cast_fp16)[name = tensor<string, []>("op_1952_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1953_cast_fp16 = softmax(axis = var_1806, x = aw_247_cast_fp16)[name = tensor<string, []>("op_1953_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1954_cast_fp16 = softmax(axis = var_1806, x = aw_249_cast_fp16)[name = tensor<string, []>("op_1954_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1955_cast_fp16 = softmax(axis = var_1806, x = aw_251_cast_fp16)[name = tensor<string, []>("op_1955_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1956_cast_fp16 = softmax(axis = var_1806, x = aw_253_cast_fp16)[name = tensor<string, []>("op_1956_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_1957_cast_fp16 = softmax(axis = var_1806, x = aw_255_cast_fp16)[name = tensor<string, []>("op_1957_cast_fp16")];
            tensor<string, []> var_1959_equation_0 = const()[name = tensor<string, []>("op_1959_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1959_cast_fp16 = einsum(equation = var_1959_equation_0, values = (var_1893_cast_fp16_0, var_1942_cast_fp16))[name = tensor<string, []>("op_1959_cast_fp16")];
            tensor<string, []> var_1961_equation_0 = const()[name = tensor<string, []>("op_1961_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1961_cast_fp16 = einsum(equation = var_1961_equation_0, values = (var_1893_cast_fp16_1, var_1943_cast_fp16))[name = tensor<string, []>("op_1961_cast_fp16")];
            tensor<string, []> var_1963_equation_0 = const()[name = tensor<string, []>("op_1963_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1963_cast_fp16 = einsum(equation = var_1963_equation_0, values = (var_1893_cast_fp16_2, var_1944_cast_fp16))[name = tensor<string, []>("op_1963_cast_fp16")];
            tensor<string, []> var_1965_equation_0 = const()[name = tensor<string, []>("op_1965_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1965_cast_fp16 = einsum(equation = var_1965_equation_0, values = (var_1893_cast_fp16_3, var_1945_cast_fp16))[name = tensor<string, []>("op_1965_cast_fp16")];
            tensor<string, []> var_1967_equation_0 = const()[name = tensor<string, []>("op_1967_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1967_cast_fp16 = einsum(equation = var_1967_equation_0, values = (var_1893_cast_fp16_4, var_1946_cast_fp16))[name = tensor<string, []>("op_1967_cast_fp16")];
            tensor<string, []> var_1969_equation_0 = const()[name = tensor<string, []>("op_1969_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1969_cast_fp16 = einsum(equation = var_1969_equation_0, values = (var_1893_cast_fp16_5, var_1947_cast_fp16))[name = tensor<string, []>("op_1969_cast_fp16")];
            tensor<string, []> var_1971_equation_0 = const()[name = tensor<string, []>("op_1971_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1971_cast_fp16 = einsum(equation = var_1971_equation_0, values = (var_1893_cast_fp16_6, var_1948_cast_fp16))[name = tensor<string, []>("op_1971_cast_fp16")];
            tensor<string, []> var_1973_equation_0 = const()[name = tensor<string, []>("op_1973_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1973_cast_fp16 = einsum(equation = var_1973_equation_0, values = (var_1893_cast_fp16_7, var_1949_cast_fp16))[name = tensor<string, []>("op_1973_cast_fp16")];
            tensor<string, []> var_1975_equation_0 = const()[name = tensor<string, []>("op_1975_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1975_cast_fp16 = einsum(equation = var_1975_equation_0, values = (var_1893_cast_fp16_8, var_1950_cast_fp16))[name = tensor<string, []>("op_1975_cast_fp16")];
            tensor<string, []> var_1977_equation_0 = const()[name = tensor<string, []>("op_1977_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1977_cast_fp16 = einsum(equation = var_1977_equation_0, values = (var_1893_cast_fp16_9, var_1951_cast_fp16))[name = tensor<string, []>("op_1977_cast_fp16")];
            tensor<string, []> var_1979_equation_0 = const()[name = tensor<string, []>("op_1979_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1979_cast_fp16 = einsum(equation = var_1979_equation_0, values = (var_1893_cast_fp16_10, var_1952_cast_fp16))[name = tensor<string, []>("op_1979_cast_fp16")];
            tensor<string, []> var_1981_equation_0 = const()[name = tensor<string, []>("op_1981_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1981_cast_fp16 = einsum(equation = var_1981_equation_0, values = (var_1893_cast_fp16_11, var_1953_cast_fp16))[name = tensor<string, []>("op_1981_cast_fp16")];
            tensor<string, []> var_1983_equation_0 = const()[name = tensor<string, []>("op_1983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1983_cast_fp16 = einsum(equation = var_1983_equation_0, values = (var_1893_cast_fp16_12, var_1954_cast_fp16))[name = tensor<string, []>("op_1983_cast_fp16")];
            tensor<string, []> var_1985_equation_0 = const()[name = tensor<string, []>("op_1985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1985_cast_fp16 = einsum(equation = var_1985_equation_0, values = (var_1893_cast_fp16_13, var_1955_cast_fp16))[name = tensor<string, []>("op_1985_cast_fp16")];
            tensor<string, []> var_1987_equation_0 = const()[name = tensor<string, []>("op_1987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1987_cast_fp16 = einsum(equation = var_1987_equation_0, values = (var_1893_cast_fp16_14, var_1956_cast_fp16))[name = tensor<string, []>("op_1987_cast_fp16")];
            tensor<string, []> var_1989_equation_0 = const()[name = tensor<string, []>("op_1989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_1989_cast_fp16 = einsum(equation = var_1989_equation_0, values = (var_1893_cast_fp16_15, var_1957_cast_fp16))[name = tensor<string, []>("op_1989_cast_fp16")];
            tensor<bool, []> input_77_interleave_0 = const()[name = tensor<string, []>("input_77_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_77_cast_fp16 = concat(axis = var_1806, interleave = input_77_interleave_0, values = (var_1959_cast_fp16, var_1961_cast_fp16, var_1963_cast_fp16, var_1965_cast_fp16, var_1967_cast_fp16, var_1969_cast_fp16, var_1971_cast_fp16, var_1973_cast_fp16, var_1975_cast_fp16, var_1977_cast_fp16, var_1979_cast_fp16, var_1981_cast_fp16, var_1983_cast_fp16, var_1985_cast_fp16, var_1987_cast_fp16, var_1989_cast_fp16))[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<string, []> var_1998_pad_type_0 = const()[name = tensor<string, []>("op_1998_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_1998_strides_0 = const()[name = tensor<string, []>("op_1998_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1998_pad_0 = const()[name = tensor<string, []>("op_1998_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1998_dilations_0 = const()[name = tensor<string, []>("op_1998_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_1998_groups_0 = const()[name = tensor<string, []>("op_1998_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_7_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47814016))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48338368))), name = tensor<string, []>("blocks_7_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_7_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48338496)))];
            tensor<fp16, [1, 1024, 1, 600]> var_1998_cast_fp16 = conv(bias = blocks_7_attn_out_bias_to_fp16, dilations = var_1998_dilations_0, groups = var_1998_groups_0, pad = var_1998_pad_0, pad_type = var_1998_pad_type_0, strides = var_1998_strides_0, weight = blocks_7_attn_out_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("op_1998_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = var_1998_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")];
            tensor<int32, [1]> input_79_axes_0 = const()[name = tensor<string, []>("input_79_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_79_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48340608)))];
            tensor<fp16, [1024]> input_79_beta_0_to_fp16 = const()[name = tensor<string, []>("input_79_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48342720)))];
            tensor<fp16, []> var_2008_to_fp16 = const()[name = tensor<string, []>("op_2008_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_79_cast_fp16 = layer_norm(axes = input_79_axes_0, beta = input_79_beta_0_to_fp16, epsilon = var_2008_to_fp16, gamma = input_79_gamma_0_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")];
            tensor<string, []> input_81_pad_type_0 = const()[name = tensor<string, []>("input_81_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_81_strides_0 = const()[name = tensor<string, []>("input_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_81_pad_0 = const()[name = tensor<string, []>("input_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_81_dilations_0 = const()[name = tensor<string, []>("input_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_81_groups_0 = const()[name = tensor<string, []>("input_81_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_7_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48344832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50442048))), name = tensor<string, []>("blocks_7_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_7_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50442176)))];
            tensor<fp16, [1, 4096, 1, 600]> input_81_cast_fp16 = conv(bias = blocks_7_mlp_0_bias_to_fp16, dilations = input_81_dilations_0, groups = input_81_groups_0, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = input_81_strides_0, weight = blocks_7_mlp_0_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<string, []> input_83_mode_0 = const()[name = tensor<string, []>("input_83_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_83_cast_fp16 = gelu(mode = input_83_mode_0, x = input_81_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
            tensor<string, []> var_2034_pad_type_0 = const()[name = tensor<string, []>("op_2034_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2034_strides_0 = const()[name = tensor<string, []>("op_2034_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2034_pad_0 = const()[name = tensor<string, []>("op_2034_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2034_dilations_0 = const()[name = tensor<string, []>("op_2034_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2034_groups_0 = const()[name = tensor<string, []>("op_2034_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_7_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50450432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52547648))), name = tensor<string, []>("blocks_7_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_7_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52547776)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2034_cast_fp16 = conv(bias = blocks_7_mlp_2_bias_to_fp16, dilations = var_2034_dilations_0, groups = var_2034_groups_0, pad = var_2034_pad_0, pad_type = var_2034_pad_type_0, strides = var_2034_strides_0, weight = blocks_7_mlp_2_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor<string, []>("op_2034_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_2034_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")];
            tensor<int32, []> var_2043 = const()[name = tensor<string, []>("op_2043"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_85_axes_0 = const()[name = tensor<string, []>("input_85_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_85_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52549888)))];
            tensor<fp16, [1024]> input_85_beta_0_to_fp16 = const()[name = tensor<string, []>("input_85_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52552000)))];
            tensor<fp16, []> var_2059_to_fp16 = const()[name = tensor<string, []>("op_2059_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_85_cast_fp16 = layer_norm(axes = input_85_axes_0, beta = input_85_beta_0_to_fp16, epsilon = var_2059_to_fp16, gamma = input_85_gamma_0_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<string, []> q_17_pad_type_0 = const()[name = tensor<string, []>("q_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_17_strides_0 = const()[name = tensor<string, []>("q_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_17_pad_0 = const()[name = tensor<string, []>("q_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_17_dilations_0 = const()[name = tensor<string, []>("q_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_17_groups_0 = const()[name = tensor<string, []>("q_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_2094_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52554112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53078464))), name = tensor<string, []>("op_2094_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_2094_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2094_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53078592)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2094_cast_fp16 = conv(bias = var_2094_bias_0_to_fp16, dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = op_2094_weight_0_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor<string, []>("op_2094_cast_fp16")];
            tensor<string, []> k_17_pad_type_0 = const()[name = tensor<string, []>("k_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_17_strides_0 = const()[name = tensor<string, []>("k_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_17_pad_0 = const()[name = tensor<string, []>("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_17_dilations_0 = const()[name = tensor<string, []>("k_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_17_groups_0 = const()[name = tensor<string, []>("k_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_8_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53080704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53605056))), name = tensor<string, []>("blocks_8_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = blocks_8_attn_key_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor<string, []>("k_17_cast_fp16")];
            tensor<string, []> var_2092_pad_type_0 = const()[name = tensor<string, []>("op_2092_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2092_strides_0 = const()[name = tensor<string, []>("op_2092_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2092_pad_0 = const()[name = tensor<string, []>("op_2092_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2092_dilations_0 = const()[name = tensor<string, []>("op_2092_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2092_groups_0 = const()[name = tensor<string, []>("op_2092_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_8_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53605184))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54129536))), name = tensor<string, []>("blocks_8_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_8_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54129664)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2092_cast_fp16 = conv(bias = blocks_8_attn_value_bias_to_fp16, dilations = var_2092_dilations_0, groups = var_2092_groups_0, pad = var_2092_pad_0, pad_type = var_2092_pad_type_0, strides = var_2092_strides_0, weight = blocks_8_attn_value_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor<string, []>("op_2092_cast_fp16")];
            tensor<int32, [16]> tile_24 = const()[name = tensor<string, []>("tile_24"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2095_axis_0 = const()[name = tensor<string, []>("op_2095_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2095_cast_fp16_15 = split(axis = var_2095_axis_0, split_sizes = tile_24, x = var_2094_cast_fp16)[name = tensor<string, []>("op_2095_cast_fp16")];
            tensor<int32, [4]> var_2112_perm_0 = const()[name = tensor<string, []>("op_2112_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_25 = const()[name = tensor<string, []>("tile_25"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2113_axis_0 = const()[name = tensor<string, []>("op_2113_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_2112_cast_fp16 = transpose(perm = var_2112_perm_0, x = k_17_cast_fp16)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_2113_cast_fp16_15 = split(axis = var_2113_axis_0, split_sizes = tile_25, x = var_2112_cast_fp16)[name = tensor<string, []>("op_2113_cast_fp16")];
            tensor<int32, [16]> tile_26 = const()[name = tensor<string, []>("tile_26"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2130_axis_0 = const()[name = tensor<string, []>("op_2130_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2130_cast_fp16_15 = split(axis = var_2130_axis_0, split_sizes = tile_26, x = var_2092_cast_fp16)[name = tensor<string, []>("op_2130_cast_fp16")];
            tensor<string, []> aw_257_equation_0 = const()[name = tensor<string, []>("aw_257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_257_cast_fp16 = einsum(equation = aw_257_equation_0, values = (var_2113_cast_fp16_0, var_2095_cast_fp16_0))[name = tensor<string, []>("aw_257_cast_fp16")];
            tensor<string, []> aw_259_equation_0 = const()[name = tensor<string, []>("aw_259_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_259_cast_fp16 = einsum(equation = aw_259_equation_0, values = (var_2113_cast_fp16_1, var_2095_cast_fp16_1))[name = tensor<string, []>("aw_259_cast_fp16")];
            tensor<string, []> aw_261_equation_0 = const()[name = tensor<string, []>("aw_261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_261_cast_fp16 = einsum(equation = aw_261_equation_0, values = (var_2113_cast_fp16_2, var_2095_cast_fp16_2))[name = tensor<string, []>("aw_261_cast_fp16")];
            tensor<string, []> aw_263_equation_0 = const()[name = tensor<string, []>("aw_263_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_263_cast_fp16 = einsum(equation = aw_263_equation_0, values = (var_2113_cast_fp16_3, var_2095_cast_fp16_3))[name = tensor<string, []>("aw_263_cast_fp16")];
            tensor<string, []> aw_265_equation_0 = const()[name = tensor<string, []>("aw_265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_265_cast_fp16 = einsum(equation = aw_265_equation_0, values = (var_2113_cast_fp16_4, var_2095_cast_fp16_4))[name = tensor<string, []>("aw_265_cast_fp16")];
            tensor<string, []> aw_267_equation_0 = const()[name = tensor<string, []>("aw_267_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_267_cast_fp16 = einsum(equation = aw_267_equation_0, values = (var_2113_cast_fp16_5, var_2095_cast_fp16_5))[name = tensor<string, []>("aw_267_cast_fp16")];
            tensor<string, []> aw_269_equation_0 = const()[name = tensor<string, []>("aw_269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_269_cast_fp16 = einsum(equation = aw_269_equation_0, values = (var_2113_cast_fp16_6, var_2095_cast_fp16_6))[name = tensor<string, []>("aw_269_cast_fp16")];
            tensor<string, []> aw_271_equation_0 = const()[name = tensor<string, []>("aw_271_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_271_cast_fp16 = einsum(equation = aw_271_equation_0, values = (var_2113_cast_fp16_7, var_2095_cast_fp16_7))[name = tensor<string, []>("aw_271_cast_fp16")];
            tensor<string, []> aw_273_equation_0 = const()[name = tensor<string, []>("aw_273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_273_cast_fp16 = einsum(equation = aw_273_equation_0, values = (var_2113_cast_fp16_8, var_2095_cast_fp16_8))[name = tensor<string, []>("aw_273_cast_fp16")];
            tensor<string, []> aw_275_equation_0 = const()[name = tensor<string, []>("aw_275_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_275_cast_fp16 = einsum(equation = aw_275_equation_0, values = (var_2113_cast_fp16_9, var_2095_cast_fp16_9))[name = tensor<string, []>("aw_275_cast_fp16")];
            tensor<string, []> aw_277_equation_0 = const()[name = tensor<string, []>("aw_277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_277_cast_fp16 = einsum(equation = aw_277_equation_0, values = (var_2113_cast_fp16_10, var_2095_cast_fp16_10))[name = tensor<string, []>("aw_277_cast_fp16")];
            tensor<string, []> aw_279_equation_0 = const()[name = tensor<string, []>("aw_279_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_279_cast_fp16 = einsum(equation = aw_279_equation_0, values = (var_2113_cast_fp16_11, var_2095_cast_fp16_11))[name = tensor<string, []>("aw_279_cast_fp16")];
            tensor<string, []> aw_281_equation_0 = const()[name = tensor<string, []>("aw_281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_281_cast_fp16 = einsum(equation = aw_281_equation_0, values = (var_2113_cast_fp16_12, var_2095_cast_fp16_12))[name = tensor<string, []>("aw_281_cast_fp16")];
            tensor<string, []> aw_283_equation_0 = const()[name = tensor<string, []>("aw_283_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_283_cast_fp16 = einsum(equation = aw_283_equation_0, values = (var_2113_cast_fp16_13, var_2095_cast_fp16_13))[name = tensor<string, []>("aw_283_cast_fp16")];
            tensor<string, []> aw_285_equation_0 = const()[name = tensor<string, []>("aw_285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_285_cast_fp16 = einsum(equation = aw_285_equation_0, values = (var_2113_cast_fp16_14, var_2095_cast_fp16_14))[name = tensor<string, []>("aw_285_cast_fp16")];
            tensor<string, []> aw_287_equation_0 = const()[name = tensor<string, []>("aw_287_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_287_cast_fp16 = einsum(equation = aw_287_equation_0, values = (var_2113_cast_fp16_15, var_2095_cast_fp16_15))[name = tensor<string, []>("aw_287_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2179_cast_fp16 = softmax(axis = var_2043, x = aw_257_cast_fp16)[name = tensor<string, []>("op_2179_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2180_cast_fp16 = softmax(axis = var_2043, x = aw_259_cast_fp16)[name = tensor<string, []>("op_2180_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2181_cast_fp16 = softmax(axis = var_2043, x = aw_261_cast_fp16)[name = tensor<string, []>("op_2181_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2182_cast_fp16 = softmax(axis = var_2043, x = aw_263_cast_fp16)[name = tensor<string, []>("op_2182_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2183_cast_fp16 = softmax(axis = var_2043, x = aw_265_cast_fp16)[name = tensor<string, []>("op_2183_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2184_cast_fp16 = softmax(axis = var_2043, x = aw_267_cast_fp16)[name = tensor<string, []>("op_2184_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2185_cast_fp16 = softmax(axis = var_2043, x = aw_269_cast_fp16)[name = tensor<string, []>("op_2185_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2186_cast_fp16 = softmax(axis = var_2043, x = aw_271_cast_fp16)[name = tensor<string, []>("op_2186_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2187_cast_fp16 = softmax(axis = var_2043, x = aw_273_cast_fp16)[name = tensor<string, []>("op_2187_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2188_cast_fp16 = softmax(axis = var_2043, x = aw_275_cast_fp16)[name = tensor<string, []>("op_2188_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2189_cast_fp16 = softmax(axis = var_2043, x = aw_277_cast_fp16)[name = tensor<string, []>("op_2189_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2190_cast_fp16 = softmax(axis = var_2043, x = aw_279_cast_fp16)[name = tensor<string, []>("op_2190_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2191_cast_fp16 = softmax(axis = var_2043, x = aw_281_cast_fp16)[name = tensor<string, []>("op_2191_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2192_cast_fp16 = softmax(axis = var_2043, x = aw_283_cast_fp16)[name = tensor<string, []>("op_2192_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2193_cast_fp16 = softmax(axis = var_2043, x = aw_285_cast_fp16)[name = tensor<string, []>("op_2193_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2194_cast_fp16 = softmax(axis = var_2043, x = aw_287_cast_fp16)[name = tensor<string, []>("op_2194_cast_fp16")];
            tensor<string, []> var_2196_equation_0 = const()[name = tensor<string, []>("op_2196_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2196_cast_fp16 = einsum(equation = var_2196_equation_0, values = (var_2130_cast_fp16_0, var_2179_cast_fp16))[name = tensor<string, []>("op_2196_cast_fp16")];
            tensor<string, []> var_2198_equation_0 = const()[name = tensor<string, []>("op_2198_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2198_cast_fp16 = einsum(equation = var_2198_equation_0, values = (var_2130_cast_fp16_1, var_2180_cast_fp16))[name = tensor<string, []>("op_2198_cast_fp16")];
            tensor<string, []> var_2200_equation_0 = const()[name = tensor<string, []>("op_2200_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2200_cast_fp16 = einsum(equation = var_2200_equation_0, values = (var_2130_cast_fp16_2, var_2181_cast_fp16))[name = tensor<string, []>("op_2200_cast_fp16")];
            tensor<string, []> var_2202_equation_0 = const()[name = tensor<string, []>("op_2202_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2202_cast_fp16 = einsum(equation = var_2202_equation_0, values = (var_2130_cast_fp16_3, var_2182_cast_fp16))[name = tensor<string, []>("op_2202_cast_fp16")];
            tensor<string, []> var_2204_equation_0 = const()[name = tensor<string, []>("op_2204_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2204_cast_fp16 = einsum(equation = var_2204_equation_0, values = (var_2130_cast_fp16_4, var_2183_cast_fp16))[name = tensor<string, []>("op_2204_cast_fp16")];
            tensor<string, []> var_2206_equation_0 = const()[name = tensor<string, []>("op_2206_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2206_cast_fp16 = einsum(equation = var_2206_equation_0, values = (var_2130_cast_fp16_5, var_2184_cast_fp16))[name = tensor<string, []>("op_2206_cast_fp16")];
            tensor<string, []> var_2208_equation_0 = const()[name = tensor<string, []>("op_2208_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2208_cast_fp16 = einsum(equation = var_2208_equation_0, values = (var_2130_cast_fp16_6, var_2185_cast_fp16))[name = tensor<string, []>("op_2208_cast_fp16")];
            tensor<string, []> var_2210_equation_0 = const()[name = tensor<string, []>("op_2210_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2210_cast_fp16 = einsum(equation = var_2210_equation_0, values = (var_2130_cast_fp16_7, var_2186_cast_fp16))[name = tensor<string, []>("op_2210_cast_fp16")];
            tensor<string, []> var_2212_equation_0 = const()[name = tensor<string, []>("op_2212_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2212_cast_fp16 = einsum(equation = var_2212_equation_0, values = (var_2130_cast_fp16_8, var_2187_cast_fp16))[name = tensor<string, []>("op_2212_cast_fp16")];
            tensor<string, []> var_2214_equation_0 = const()[name = tensor<string, []>("op_2214_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2214_cast_fp16 = einsum(equation = var_2214_equation_0, values = (var_2130_cast_fp16_9, var_2188_cast_fp16))[name = tensor<string, []>("op_2214_cast_fp16")];
            tensor<string, []> var_2216_equation_0 = const()[name = tensor<string, []>("op_2216_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2216_cast_fp16 = einsum(equation = var_2216_equation_0, values = (var_2130_cast_fp16_10, var_2189_cast_fp16))[name = tensor<string, []>("op_2216_cast_fp16")];
            tensor<string, []> var_2218_equation_0 = const()[name = tensor<string, []>("op_2218_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2218_cast_fp16 = einsum(equation = var_2218_equation_0, values = (var_2130_cast_fp16_11, var_2190_cast_fp16))[name = tensor<string, []>("op_2218_cast_fp16")];
            tensor<string, []> var_2220_equation_0 = const()[name = tensor<string, []>("op_2220_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2220_cast_fp16 = einsum(equation = var_2220_equation_0, values = (var_2130_cast_fp16_12, var_2191_cast_fp16))[name = tensor<string, []>("op_2220_cast_fp16")];
            tensor<string, []> var_2222_equation_0 = const()[name = tensor<string, []>("op_2222_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2222_cast_fp16 = einsum(equation = var_2222_equation_0, values = (var_2130_cast_fp16_13, var_2192_cast_fp16))[name = tensor<string, []>("op_2222_cast_fp16")];
            tensor<string, []> var_2224_equation_0 = const()[name = tensor<string, []>("op_2224_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2224_cast_fp16 = einsum(equation = var_2224_equation_0, values = (var_2130_cast_fp16_14, var_2193_cast_fp16))[name = tensor<string, []>("op_2224_cast_fp16")];
            tensor<string, []> var_2226_equation_0 = const()[name = tensor<string, []>("op_2226_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2226_cast_fp16 = einsum(equation = var_2226_equation_0, values = (var_2130_cast_fp16_15, var_2194_cast_fp16))[name = tensor<string, []>("op_2226_cast_fp16")];
            tensor<bool, []> input_87_interleave_0 = const()[name = tensor<string, []>("input_87_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_87_cast_fp16 = concat(axis = var_2043, interleave = input_87_interleave_0, values = (var_2196_cast_fp16, var_2198_cast_fp16, var_2200_cast_fp16, var_2202_cast_fp16, var_2204_cast_fp16, var_2206_cast_fp16, var_2208_cast_fp16, var_2210_cast_fp16, var_2212_cast_fp16, var_2214_cast_fp16, var_2216_cast_fp16, var_2218_cast_fp16, var_2220_cast_fp16, var_2222_cast_fp16, var_2224_cast_fp16, var_2226_cast_fp16))[name = tensor<string, []>("input_87_cast_fp16")];
            tensor<string, []> var_2235_pad_type_0 = const()[name = tensor<string, []>("op_2235_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2235_strides_0 = const()[name = tensor<string, []>("op_2235_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2235_pad_0 = const()[name = tensor<string, []>("op_2235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2235_dilations_0 = const()[name = tensor<string, []>("op_2235_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2235_groups_0 = const()[name = tensor<string, []>("op_2235_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_8_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54131776))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54656128))), name = tensor<string, []>("blocks_8_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_8_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54656256)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2235_cast_fp16 = conv(bias = blocks_8_attn_out_bias_to_fp16, dilations = var_2235_dilations_0, groups = var_2235_groups_0, pad = var_2235_pad_0, pad_type = var_2235_pad_type_0, strides = var_2235_strides_0, weight = blocks_8_attn_out_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor<string, []>("op_2235_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = var_2235_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")];
            tensor<int32, [1]> input_89_axes_0 = const()[name = tensor<string, []>("input_89_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_89_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_89_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54658368)))];
            tensor<fp16, [1024]> input_89_beta_0_to_fp16 = const()[name = tensor<string, []>("input_89_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54660480)))];
            tensor<fp16, []> var_2245_to_fp16 = const()[name = tensor<string, []>("op_2245_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_89_cast_fp16 = layer_norm(axes = input_89_axes_0, beta = input_89_beta_0_to_fp16, epsilon = var_2245_to_fp16, gamma = input_89_gamma_0_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")];
            tensor<string, []> input_91_pad_type_0 = const()[name = tensor<string, []>("input_91_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_91_strides_0 = const()[name = tensor<string, []>("input_91_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_91_pad_0 = const()[name = tensor<string, []>("input_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_91_dilations_0 = const()[name = tensor<string, []>("input_91_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_91_groups_0 = const()[name = tensor<string, []>("input_91_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_8_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54662592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56759808))), name = tensor<string, []>("blocks_8_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_8_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56759936)))];
            tensor<fp16, [1, 4096, 1, 600]> input_91_cast_fp16 = conv(bias = blocks_8_mlp_0_bias_to_fp16, dilations = input_91_dilations_0, groups = input_91_groups_0, pad = input_91_pad_0, pad_type = input_91_pad_type_0, strides = input_91_strides_0, weight = blocks_8_mlp_0_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<string, []> input_93_mode_0 = const()[name = tensor<string, []>("input_93_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_93_cast_fp16 = gelu(mode = input_93_mode_0, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")];
            tensor<string, []> var_2271_pad_type_0 = const()[name = tensor<string, []>("op_2271_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2271_strides_0 = const()[name = tensor<string, []>("op_2271_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2271_pad_0 = const()[name = tensor<string, []>("op_2271_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2271_dilations_0 = const()[name = tensor<string, []>("op_2271_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2271_groups_0 = const()[name = tensor<string, []>("op_2271_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_8_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56768192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58865408))), name = tensor<string, []>("blocks_8_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_8_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58865536)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2271_cast_fp16 = conv(bias = blocks_8_mlp_2_bias_to_fp16, dilations = var_2271_dilations_0, groups = var_2271_groups_0, pad = var_2271_pad_0, pad_type = var_2271_pad_type_0, strides = var_2271_strides_0, weight = blocks_8_mlp_2_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor<string, []>("op_2271_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = var_2271_cast_fp16)[name = tensor<string, []>("inputs_37_cast_fp16")];
            tensor<int32, []> var_2280 = const()[name = tensor<string, []>("op_2280"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_95_axes_0 = const()[name = tensor<string, []>("input_95_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_95_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_95_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58867648)))];
            tensor<fp16, [1024]> input_95_beta_0_to_fp16 = const()[name = tensor<string, []>("input_95_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58869760)))];
            tensor<fp16, []> var_2296_to_fp16 = const()[name = tensor<string, []>("op_2296_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_95_cast_fp16 = layer_norm(axes = input_95_axes_0, beta = input_95_beta_0_to_fp16, epsilon = var_2296_to_fp16, gamma = input_95_gamma_0_to_fp16, x = inputs_37_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<string, []> q_19_pad_type_0 = const()[name = tensor<string, []>("q_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_19_strides_0 = const()[name = tensor<string, []>("q_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_19_pad_0 = const()[name = tensor<string, []>("q_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_19_dilations_0 = const()[name = tensor<string, []>("q_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_19_groups_0 = const()[name = tensor<string, []>("q_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_2331_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58871872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59396224))), name = tensor<string, []>("op_2331_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_2331_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2331_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59396352)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2331_cast_fp16 = conv(bias = var_2331_bias_0_to_fp16, dilations = q_19_dilations_0, groups = q_19_groups_0, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = q_19_strides_0, weight = op_2331_weight_0_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor<string, []>("op_2331_cast_fp16")];
            tensor<string, []> k_19_pad_type_0 = const()[name = tensor<string, []>("k_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_19_strides_0 = const()[name = tensor<string, []>("k_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_19_pad_0 = const()[name = tensor<string, []>("k_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_19_dilations_0 = const()[name = tensor<string, []>("k_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_19_groups_0 = const()[name = tensor<string, []>("k_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_9_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59398464))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59922816))), name = tensor<string, []>("blocks_9_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_19_cast_fp16 = conv(dilations = k_19_dilations_0, groups = k_19_groups_0, pad = k_19_pad_0, pad_type = k_19_pad_type_0, strides = k_19_strides_0, weight = blocks_9_attn_key_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor<string, []>("k_19_cast_fp16")];
            tensor<string, []> var_2329_pad_type_0 = const()[name = tensor<string, []>("op_2329_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2329_strides_0 = const()[name = tensor<string, []>("op_2329_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2329_pad_0 = const()[name = tensor<string, []>("op_2329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2329_dilations_0 = const()[name = tensor<string, []>("op_2329_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2329_groups_0 = const()[name = tensor<string, []>("op_2329_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_9_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59922944))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60447296))), name = tensor<string, []>("blocks_9_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_9_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60447424)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2329_cast_fp16 = conv(bias = blocks_9_attn_value_bias_to_fp16, dilations = var_2329_dilations_0, groups = var_2329_groups_0, pad = var_2329_pad_0, pad_type = var_2329_pad_type_0, strides = var_2329_strides_0, weight = blocks_9_attn_value_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor<string, []>("op_2329_cast_fp16")];
            tensor<int32, [16]> tile_27 = const()[name = tensor<string, []>("tile_27"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2332_axis_0 = const()[name = tensor<string, []>("op_2332_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2332_cast_fp16_15 = split(axis = var_2332_axis_0, split_sizes = tile_27, x = var_2331_cast_fp16)[name = tensor<string, []>("op_2332_cast_fp16")];
            tensor<int32, [4]> var_2349_perm_0 = const()[name = tensor<string, []>("op_2349_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_28 = const()[name = tensor<string, []>("tile_28"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2350_axis_0 = const()[name = tensor<string, []>("op_2350_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_2349_cast_fp16 = transpose(perm = var_2349_perm_0, x = k_19_cast_fp16)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_2350_cast_fp16_15 = split(axis = var_2350_axis_0, split_sizes = tile_28, x = var_2349_cast_fp16)[name = tensor<string, []>("op_2350_cast_fp16")];
            tensor<int32, [16]> tile_29 = const()[name = tensor<string, []>("tile_29"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2367_axis_0 = const()[name = tensor<string, []>("op_2367_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2367_cast_fp16_15 = split(axis = var_2367_axis_0, split_sizes = tile_29, x = var_2329_cast_fp16)[name = tensor<string, []>("op_2367_cast_fp16")];
            tensor<string, []> aw_289_equation_0 = const()[name = tensor<string, []>("aw_289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_289_cast_fp16 = einsum(equation = aw_289_equation_0, values = (var_2350_cast_fp16_0, var_2332_cast_fp16_0))[name = tensor<string, []>("aw_289_cast_fp16")];
            tensor<string, []> aw_291_equation_0 = const()[name = tensor<string, []>("aw_291_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_291_cast_fp16 = einsum(equation = aw_291_equation_0, values = (var_2350_cast_fp16_1, var_2332_cast_fp16_1))[name = tensor<string, []>("aw_291_cast_fp16")];
            tensor<string, []> aw_293_equation_0 = const()[name = tensor<string, []>("aw_293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_293_cast_fp16 = einsum(equation = aw_293_equation_0, values = (var_2350_cast_fp16_2, var_2332_cast_fp16_2))[name = tensor<string, []>("aw_293_cast_fp16")];
            tensor<string, []> aw_295_equation_0 = const()[name = tensor<string, []>("aw_295_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_295_cast_fp16 = einsum(equation = aw_295_equation_0, values = (var_2350_cast_fp16_3, var_2332_cast_fp16_3))[name = tensor<string, []>("aw_295_cast_fp16")];
            tensor<string, []> aw_297_equation_0 = const()[name = tensor<string, []>("aw_297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_297_cast_fp16 = einsum(equation = aw_297_equation_0, values = (var_2350_cast_fp16_4, var_2332_cast_fp16_4))[name = tensor<string, []>("aw_297_cast_fp16")];
            tensor<string, []> aw_299_equation_0 = const()[name = tensor<string, []>("aw_299_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_299_cast_fp16 = einsum(equation = aw_299_equation_0, values = (var_2350_cast_fp16_5, var_2332_cast_fp16_5))[name = tensor<string, []>("aw_299_cast_fp16")];
            tensor<string, []> aw_301_equation_0 = const()[name = tensor<string, []>("aw_301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_301_cast_fp16 = einsum(equation = aw_301_equation_0, values = (var_2350_cast_fp16_6, var_2332_cast_fp16_6))[name = tensor<string, []>("aw_301_cast_fp16")];
            tensor<string, []> aw_303_equation_0 = const()[name = tensor<string, []>("aw_303_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_303_cast_fp16 = einsum(equation = aw_303_equation_0, values = (var_2350_cast_fp16_7, var_2332_cast_fp16_7))[name = tensor<string, []>("aw_303_cast_fp16")];
            tensor<string, []> aw_305_equation_0 = const()[name = tensor<string, []>("aw_305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_305_cast_fp16 = einsum(equation = aw_305_equation_0, values = (var_2350_cast_fp16_8, var_2332_cast_fp16_8))[name = tensor<string, []>("aw_305_cast_fp16")];
            tensor<string, []> aw_307_equation_0 = const()[name = tensor<string, []>("aw_307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_307_cast_fp16 = einsum(equation = aw_307_equation_0, values = (var_2350_cast_fp16_9, var_2332_cast_fp16_9))[name = tensor<string, []>("aw_307_cast_fp16")];
            tensor<string, []> aw_309_equation_0 = const()[name = tensor<string, []>("aw_309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_309_cast_fp16 = einsum(equation = aw_309_equation_0, values = (var_2350_cast_fp16_10, var_2332_cast_fp16_10))[name = tensor<string, []>("aw_309_cast_fp16")];
            tensor<string, []> aw_311_equation_0 = const()[name = tensor<string, []>("aw_311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_311_cast_fp16 = einsum(equation = aw_311_equation_0, values = (var_2350_cast_fp16_11, var_2332_cast_fp16_11))[name = tensor<string, []>("aw_311_cast_fp16")];
            tensor<string, []> aw_313_equation_0 = const()[name = tensor<string, []>("aw_313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_313_cast_fp16 = einsum(equation = aw_313_equation_0, values = (var_2350_cast_fp16_12, var_2332_cast_fp16_12))[name = tensor<string, []>("aw_313_cast_fp16")];
            tensor<string, []> aw_315_equation_0 = const()[name = tensor<string, []>("aw_315_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_315_cast_fp16 = einsum(equation = aw_315_equation_0, values = (var_2350_cast_fp16_13, var_2332_cast_fp16_13))[name = tensor<string, []>("aw_315_cast_fp16")];
            tensor<string, []> aw_317_equation_0 = const()[name = tensor<string, []>("aw_317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_317_cast_fp16 = einsum(equation = aw_317_equation_0, values = (var_2350_cast_fp16_14, var_2332_cast_fp16_14))[name = tensor<string, []>("aw_317_cast_fp16")];
            tensor<string, []> aw_319_equation_0 = const()[name = tensor<string, []>("aw_319_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_319_cast_fp16 = einsum(equation = aw_319_equation_0, values = (var_2350_cast_fp16_15, var_2332_cast_fp16_15))[name = tensor<string, []>("aw_319_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2416_cast_fp16 = softmax(axis = var_2280, x = aw_289_cast_fp16)[name = tensor<string, []>("op_2416_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2417_cast_fp16 = softmax(axis = var_2280, x = aw_291_cast_fp16)[name = tensor<string, []>("op_2417_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2418_cast_fp16 = softmax(axis = var_2280, x = aw_293_cast_fp16)[name = tensor<string, []>("op_2418_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2419_cast_fp16 = softmax(axis = var_2280, x = aw_295_cast_fp16)[name = tensor<string, []>("op_2419_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2420_cast_fp16 = softmax(axis = var_2280, x = aw_297_cast_fp16)[name = tensor<string, []>("op_2420_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2421_cast_fp16 = softmax(axis = var_2280, x = aw_299_cast_fp16)[name = tensor<string, []>("op_2421_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2422_cast_fp16 = softmax(axis = var_2280, x = aw_301_cast_fp16)[name = tensor<string, []>("op_2422_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2423_cast_fp16 = softmax(axis = var_2280, x = aw_303_cast_fp16)[name = tensor<string, []>("op_2423_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2424_cast_fp16 = softmax(axis = var_2280, x = aw_305_cast_fp16)[name = tensor<string, []>("op_2424_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2425_cast_fp16 = softmax(axis = var_2280, x = aw_307_cast_fp16)[name = tensor<string, []>("op_2425_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2426_cast_fp16 = softmax(axis = var_2280, x = aw_309_cast_fp16)[name = tensor<string, []>("op_2426_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2427_cast_fp16 = softmax(axis = var_2280, x = aw_311_cast_fp16)[name = tensor<string, []>("op_2427_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2428_cast_fp16 = softmax(axis = var_2280, x = aw_313_cast_fp16)[name = tensor<string, []>("op_2428_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2429_cast_fp16 = softmax(axis = var_2280, x = aw_315_cast_fp16)[name = tensor<string, []>("op_2429_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2430_cast_fp16 = softmax(axis = var_2280, x = aw_317_cast_fp16)[name = tensor<string, []>("op_2430_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2431_cast_fp16 = softmax(axis = var_2280, x = aw_319_cast_fp16)[name = tensor<string, []>("op_2431_cast_fp16")];
            tensor<string, []> var_2433_equation_0 = const()[name = tensor<string, []>("op_2433_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2433_cast_fp16 = einsum(equation = var_2433_equation_0, values = (var_2367_cast_fp16_0, var_2416_cast_fp16))[name = tensor<string, []>("op_2433_cast_fp16")];
            tensor<string, []> var_2435_equation_0 = const()[name = tensor<string, []>("op_2435_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2435_cast_fp16 = einsum(equation = var_2435_equation_0, values = (var_2367_cast_fp16_1, var_2417_cast_fp16))[name = tensor<string, []>("op_2435_cast_fp16")];
            tensor<string, []> var_2437_equation_0 = const()[name = tensor<string, []>("op_2437_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2437_cast_fp16 = einsum(equation = var_2437_equation_0, values = (var_2367_cast_fp16_2, var_2418_cast_fp16))[name = tensor<string, []>("op_2437_cast_fp16")];
            tensor<string, []> var_2439_equation_0 = const()[name = tensor<string, []>("op_2439_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2439_cast_fp16 = einsum(equation = var_2439_equation_0, values = (var_2367_cast_fp16_3, var_2419_cast_fp16))[name = tensor<string, []>("op_2439_cast_fp16")];
            tensor<string, []> var_2441_equation_0 = const()[name = tensor<string, []>("op_2441_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2441_cast_fp16 = einsum(equation = var_2441_equation_0, values = (var_2367_cast_fp16_4, var_2420_cast_fp16))[name = tensor<string, []>("op_2441_cast_fp16")];
            tensor<string, []> var_2443_equation_0 = const()[name = tensor<string, []>("op_2443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2443_cast_fp16 = einsum(equation = var_2443_equation_0, values = (var_2367_cast_fp16_5, var_2421_cast_fp16))[name = tensor<string, []>("op_2443_cast_fp16")];
            tensor<string, []> var_2445_equation_0 = const()[name = tensor<string, []>("op_2445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2445_cast_fp16 = einsum(equation = var_2445_equation_0, values = (var_2367_cast_fp16_6, var_2422_cast_fp16))[name = tensor<string, []>("op_2445_cast_fp16")];
            tensor<string, []> var_2447_equation_0 = const()[name = tensor<string, []>("op_2447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2447_cast_fp16 = einsum(equation = var_2447_equation_0, values = (var_2367_cast_fp16_7, var_2423_cast_fp16))[name = tensor<string, []>("op_2447_cast_fp16")];
            tensor<string, []> var_2449_equation_0 = const()[name = tensor<string, []>("op_2449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2449_cast_fp16 = einsum(equation = var_2449_equation_0, values = (var_2367_cast_fp16_8, var_2424_cast_fp16))[name = tensor<string, []>("op_2449_cast_fp16")];
            tensor<string, []> var_2451_equation_0 = const()[name = tensor<string, []>("op_2451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2451_cast_fp16 = einsum(equation = var_2451_equation_0, values = (var_2367_cast_fp16_9, var_2425_cast_fp16))[name = tensor<string, []>("op_2451_cast_fp16")];
            tensor<string, []> var_2453_equation_0 = const()[name = tensor<string, []>("op_2453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2453_cast_fp16 = einsum(equation = var_2453_equation_0, values = (var_2367_cast_fp16_10, var_2426_cast_fp16))[name = tensor<string, []>("op_2453_cast_fp16")];
            tensor<string, []> var_2455_equation_0 = const()[name = tensor<string, []>("op_2455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2455_cast_fp16 = einsum(equation = var_2455_equation_0, values = (var_2367_cast_fp16_11, var_2427_cast_fp16))[name = tensor<string, []>("op_2455_cast_fp16")];
            tensor<string, []> var_2457_equation_0 = const()[name = tensor<string, []>("op_2457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2457_cast_fp16 = einsum(equation = var_2457_equation_0, values = (var_2367_cast_fp16_12, var_2428_cast_fp16))[name = tensor<string, []>("op_2457_cast_fp16")];
            tensor<string, []> var_2459_equation_0 = const()[name = tensor<string, []>("op_2459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2459_cast_fp16 = einsum(equation = var_2459_equation_0, values = (var_2367_cast_fp16_13, var_2429_cast_fp16))[name = tensor<string, []>("op_2459_cast_fp16")];
            tensor<string, []> var_2461_equation_0 = const()[name = tensor<string, []>("op_2461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2461_cast_fp16 = einsum(equation = var_2461_equation_0, values = (var_2367_cast_fp16_14, var_2430_cast_fp16))[name = tensor<string, []>("op_2461_cast_fp16")];
            tensor<string, []> var_2463_equation_0 = const()[name = tensor<string, []>("op_2463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2463_cast_fp16 = einsum(equation = var_2463_equation_0, values = (var_2367_cast_fp16_15, var_2431_cast_fp16))[name = tensor<string, []>("op_2463_cast_fp16")];
            tensor<bool, []> input_97_interleave_0 = const()[name = tensor<string, []>("input_97_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_97_cast_fp16 = concat(axis = var_2280, interleave = input_97_interleave_0, values = (var_2433_cast_fp16, var_2435_cast_fp16, var_2437_cast_fp16, var_2439_cast_fp16, var_2441_cast_fp16, var_2443_cast_fp16, var_2445_cast_fp16, var_2447_cast_fp16, var_2449_cast_fp16, var_2451_cast_fp16, var_2453_cast_fp16, var_2455_cast_fp16, var_2457_cast_fp16, var_2459_cast_fp16, var_2461_cast_fp16, var_2463_cast_fp16))[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<string, []> var_2472_pad_type_0 = const()[name = tensor<string, []>("op_2472_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2472_strides_0 = const()[name = tensor<string, []>("op_2472_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2472_pad_0 = const()[name = tensor<string, []>("op_2472_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2472_dilations_0 = const()[name = tensor<string, []>("op_2472_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2472_groups_0 = const()[name = tensor<string, []>("op_2472_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_9_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60449536))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60973888))), name = tensor<string, []>("blocks_9_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_9_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60974016)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2472_cast_fp16 = conv(bias = blocks_9_attn_out_bias_to_fp16, dilations = var_2472_dilations_0, groups = var_2472_groups_0, pad = var_2472_pad_0, pad_type = var_2472_pad_type_0, strides = var_2472_strides_0, weight = blocks_9_attn_out_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor<string, []>("op_2472_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = var_2472_cast_fp16)[name = tensor<string, []>("inputs_39_cast_fp16")];
            tensor<int32, [1]> input_99_axes_0 = const()[name = tensor<string, []>("input_99_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_99_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_99_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60976128)))];
            tensor<fp16, [1024]> input_99_beta_0_to_fp16 = const()[name = tensor<string, []>("input_99_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60978240)))];
            tensor<fp16, []> var_2482_to_fp16 = const()[name = tensor<string, []>("op_2482_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_99_cast_fp16 = layer_norm(axes = input_99_axes_0, beta = input_99_beta_0_to_fp16, epsilon = var_2482_to_fp16, gamma = input_99_gamma_0_to_fp16, x = inputs_39_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
            tensor<string, []> input_101_pad_type_0 = const()[name = tensor<string, []>("input_101_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_101_strides_0 = const()[name = tensor<string, []>("input_101_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_101_pad_0 = const()[name = tensor<string, []>("input_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_101_dilations_0 = const()[name = tensor<string, []>("input_101_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_101_groups_0 = const()[name = tensor<string, []>("input_101_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_9_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60980352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63077568))), name = tensor<string, []>("blocks_9_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_9_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63077696)))];
            tensor<fp16, [1, 4096, 1, 600]> input_101_cast_fp16 = conv(bias = blocks_9_mlp_0_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = blocks_9_mlp_0_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")];
            tensor<string, []> input_103_mode_0 = const()[name = tensor<string, []>("input_103_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<string, []> var_2508_pad_type_0 = const()[name = tensor<string, []>("op_2508_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2508_strides_0 = const()[name = tensor<string, []>("op_2508_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2508_pad_0 = const()[name = tensor<string, []>("op_2508_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2508_dilations_0 = const()[name = tensor<string, []>("op_2508_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2508_groups_0 = const()[name = tensor<string, []>("op_2508_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_9_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63085952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65183168))), name = tensor<string, []>("blocks_9_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_9_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65183296)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2508_cast_fp16 = conv(bias = blocks_9_mlp_2_bias_to_fp16, dilations = var_2508_dilations_0, groups = var_2508_groups_0, pad = var_2508_pad_0, pad_type = var_2508_pad_type_0, strides = var_2508_strides_0, weight = blocks_9_mlp_2_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor<string, []>("op_2508_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = var_2508_cast_fp16)[name = tensor<string, []>("inputs_41_cast_fp16")];
            tensor<int32, []> var_2517 = const()[name = tensor<string, []>("op_2517"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_105_axes_0 = const()[name = tensor<string, []>("input_105_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_105_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_105_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65185408)))];
            tensor<fp16, [1024]> input_105_beta_0_to_fp16 = const()[name = tensor<string, []>("input_105_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65187520)))];
            tensor<fp16, []> var_2533_to_fp16 = const()[name = tensor<string, []>("op_2533_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_105_cast_fp16 = layer_norm(axes = input_105_axes_0, beta = input_105_beta_0_to_fp16, epsilon = var_2533_to_fp16, gamma = input_105_gamma_0_to_fp16, x = inputs_41_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<string, []> q_21_pad_type_0 = const()[name = tensor<string, []>("q_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_21_strides_0 = const()[name = tensor<string, []>("q_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_21_pad_0 = const()[name = tensor<string, []>("q_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_21_dilations_0 = const()[name = tensor<string, []>("q_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_21_groups_0 = const()[name = tensor<string, []>("q_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_2568_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65189632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65713984))), name = tensor<string, []>("op_2568_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_2568_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2568_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65714112)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2568_cast_fp16 = conv(bias = var_2568_bias_0_to_fp16, dilations = q_21_dilations_0, groups = q_21_groups_0, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = q_21_strides_0, weight = op_2568_weight_0_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("op_2568_cast_fp16")];
            tensor<string, []> k_21_pad_type_0 = const()[name = tensor<string, []>("k_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_21_strides_0 = const()[name = tensor<string, []>("k_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_21_pad_0 = const()[name = tensor<string, []>("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_21_dilations_0 = const()[name = tensor<string, []>("k_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_21_groups_0 = const()[name = tensor<string, []>("k_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_10_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65716224))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66240576))), name = tensor<string, []>("blocks_10_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = blocks_10_attn_key_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("k_21_cast_fp16")];
            tensor<string, []> var_2566_pad_type_0 = const()[name = tensor<string, []>("op_2566_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2566_strides_0 = const()[name = tensor<string, []>("op_2566_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2566_pad_0 = const()[name = tensor<string, []>("op_2566_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2566_dilations_0 = const()[name = tensor<string, []>("op_2566_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2566_groups_0 = const()[name = tensor<string, []>("op_2566_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_10_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66240704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66765056))), name = tensor<string, []>("blocks_10_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_10_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66765184)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2566_cast_fp16 = conv(bias = blocks_10_attn_value_bias_to_fp16, dilations = var_2566_dilations_0, groups = var_2566_groups_0, pad = var_2566_pad_0, pad_type = var_2566_pad_type_0, strides = var_2566_strides_0, weight = blocks_10_attn_value_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("op_2566_cast_fp16")];
            tensor<int32, [16]> tile_30 = const()[name = tensor<string, []>("tile_30"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2569_axis_0 = const()[name = tensor<string, []>("op_2569_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2569_cast_fp16_15 = split(axis = var_2569_axis_0, split_sizes = tile_30, x = var_2568_cast_fp16)[name = tensor<string, []>("op_2569_cast_fp16")];
            tensor<int32, [4]> var_2586_perm_0 = const()[name = tensor<string, []>("op_2586_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_31 = const()[name = tensor<string, []>("tile_31"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2587_axis_0 = const()[name = tensor<string, []>("op_2587_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_2586_cast_fp16 = transpose(perm = var_2586_perm_0, x = k_21_cast_fp16)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_2587_cast_fp16_15 = split(axis = var_2587_axis_0, split_sizes = tile_31, x = var_2586_cast_fp16)[name = tensor<string, []>("op_2587_cast_fp16")];
            tensor<int32, [16]> tile_32 = const()[name = tensor<string, []>("tile_32"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2604_axis_0 = const()[name = tensor<string, []>("op_2604_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2604_cast_fp16_15 = split(axis = var_2604_axis_0, split_sizes = tile_32, x = var_2566_cast_fp16)[name = tensor<string, []>("op_2604_cast_fp16")];
            tensor<string, []> aw_321_equation_0 = const()[name = tensor<string, []>("aw_321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_321_cast_fp16 = einsum(equation = aw_321_equation_0, values = (var_2587_cast_fp16_0, var_2569_cast_fp16_0))[name = tensor<string, []>("aw_321_cast_fp16")];
            tensor<string, []> aw_323_equation_0 = const()[name = tensor<string, []>("aw_323_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_323_cast_fp16 = einsum(equation = aw_323_equation_0, values = (var_2587_cast_fp16_1, var_2569_cast_fp16_1))[name = tensor<string, []>("aw_323_cast_fp16")];
            tensor<string, []> aw_325_equation_0 = const()[name = tensor<string, []>("aw_325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_325_cast_fp16 = einsum(equation = aw_325_equation_0, values = (var_2587_cast_fp16_2, var_2569_cast_fp16_2))[name = tensor<string, []>("aw_325_cast_fp16")];
            tensor<string, []> aw_327_equation_0 = const()[name = tensor<string, []>("aw_327_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_327_cast_fp16 = einsum(equation = aw_327_equation_0, values = (var_2587_cast_fp16_3, var_2569_cast_fp16_3))[name = tensor<string, []>("aw_327_cast_fp16")];
            tensor<string, []> aw_329_equation_0 = const()[name = tensor<string, []>("aw_329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_329_cast_fp16 = einsum(equation = aw_329_equation_0, values = (var_2587_cast_fp16_4, var_2569_cast_fp16_4))[name = tensor<string, []>("aw_329_cast_fp16")];
            tensor<string, []> aw_331_equation_0 = const()[name = tensor<string, []>("aw_331_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_331_cast_fp16 = einsum(equation = aw_331_equation_0, values = (var_2587_cast_fp16_5, var_2569_cast_fp16_5))[name = tensor<string, []>("aw_331_cast_fp16")];
            tensor<string, []> aw_333_equation_0 = const()[name = tensor<string, []>("aw_333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_333_cast_fp16 = einsum(equation = aw_333_equation_0, values = (var_2587_cast_fp16_6, var_2569_cast_fp16_6))[name = tensor<string, []>("aw_333_cast_fp16")];
            tensor<string, []> aw_335_equation_0 = const()[name = tensor<string, []>("aw_335_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_335_cast_fp16 = einsum(equation = aw_335_equation_0, values = (var_2587_cast_fp16_7, var_2569_cast_fp16_7))[name = tensor<string, []>("aw_335_cast_fp16")];
            tensor<string, []> aw_337_equation_0 = const()[name = tensor<string, []>("aw_337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_337_cast_fp16 = einsum(equation = aw_337_equation_0, values = (var_2587_cast_fp16_8, var_2569_cast_fp16_8))[name = tensor<string, []>("aw_337_cast_fp16")];
            tensor<string, []> aw_339_equation_0 = const()[name = tensor<string, []>("aw_339_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_339_cast_fp16 = einsum(equation = aw_339_equation_0, values = (var_2587_cast_fp16_9, var_2569_cast_fp16_9))[name = tensor<string, []>("aw_339_cast_fp16")];
            tensor<string, []> aw_341_equation_0 = const()[name = tensor<string, []>("aw_341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_341_cast_fp16 = einsum(equation = aw_341_equation_0, values = (var_2587_cast_fp16_10, var_2569_cast_fp16_10))[name = tensor<string, []>("aw_341_cast_fp16")];
            tensor<string, []> aw_343_equation_0 = const()[name = tensor<string, []>("aw_343_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_343_cast_fp16 = einsum(equation = aw_343_equation_0, values = (var_2587_cast_fp16_11, var_2569_cast_fp16_11))[name = tensor<string, []>("aw_343_cast_fp16")];
            tensor<string, []> aw_345_equation_0 = const()[name = tensor<string, []>("aw_345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_345_cast_fp16 = einsum(equation = aw_345_equation_0, values = (var_2587_cast_fp16_12, var_2569_cast_fp16_12))[name = tensor<string, []>("aw_345_cast_fp16")];
            tensor<string, []> aw_347_equation_0 = const()[name = tensor<string, []>("aw_347_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_347_cast_fp16 = einsum(equation = aw_347_equation_0, values = (var_2587_cast_fp16_13, var_2569_cast_fp16_13))[name = tensor<string, []>("aw_347_cast_fp16")];
            tensor<string, []> aw_349_equation_0 = const()[name = tensor<string, []>("aw_349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_349_cast_fp16 = einsum(equation = aw_349_equation_0, values = (var_2587_cast_fp16_14, var_2569_cast_fp16_14))[name = tensor<string, []>("aw_349_cast_fp16")];
            tensor<string, []> aw_351_equation_0 = const()[name = tensor<string, []>("aw_351_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_351_cast_fp16 = einsum(equation = aw_351_equation_0, values = (var_2587_cast_fp16_15, var_2569_cast_fp16_15))[name = tensor<string, []>("aw_351_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2653_cast_fp16 = softmax(axis = var_2517, x = aw_321_cast_fp16)[name = tensor<string, []>("op_2653_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2654_cast_fp16 = softmax(axis = var_2517, x = aw_323_cast_fp16)[name = tensor<string, []>("op_2654_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2655_cast_fp16 = softmax(axis = var_2517, x = aw_325_cast_fp16)[name = tensor<string, []>("op_2655_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2656_cast_fp16 = softmax(axis = var_2517, x = aw_327_cast_fp16)[name = tensor<string, []>("op_2656_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2657_cast_fp16 = softmax(axis = var_2517, x = aw_329_cast_fp16)[name = tensor<string, []>("op_2657_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2658_cast_fp16 = softmax(axis = var_2517, x = aw_331_cast_fp16)[name = tensor<string, []>("op_2658_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2659_cast_fp16 = softmax(axis = var_2517, x = aw_333_cast_fp16)[name = tensor<string, []>("op_2659_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2660_cast_fp16 = softmax(axis = var_2517, x = aw_335_cast_fp16)[name = tensor<string, []>("op_2660_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2661_cast_fp16 = softmax(axis = var_2517, x = aw_337_cast_fp16)[name = tensor<string, []>("op_2661_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2662_cast_fp16 = softmax(axis = var_2517, x = aw_339_cast_fp16)[name = tensor<string, []>("op_2662_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2663_cast_fp16 = softmax(axis = var_2517, x = aw_341_cast_fp16)[name = tensor<string, []>("op_2663_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2664_cast_fp16 = softmax(axis = var_2517, x = aw_343_cast_fp16)[name = tensor<string, []>("op_2664_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2665_cast_fp16 = softmax(axis = var_2517, x = aw_345_cast_fp16)[name = tensor<string, []>("op_2665_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2666_cast_fp16 = softmax(axis = var_2517, x = aw_347_cast_fp16)[name = tensor<string, []>("op_2666_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2667_cast_fp16 = softmax(axis = var_2517, x = aw_349_cast_fp16)[name = tensor<string, []>("op_2667_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2668_cast_fp16 = softmax(axis = var_2517, x = aw_351_cast_fp16)[name = tensor<string, []>("op_2668_cast_fp16")];
            tensor<string, []> var_2670_equation_0 = const()[name = tensor<string, []>("op_2670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2670_cast_fp16 = einsum(equation = var_2670_equation_0, values = (var_2604_cast_fp16_0, var_2653_cast_fp16))[name = tensor<string, []>("op_2670_cast_fp16")];
            tensor<string, []> var_2672_equation_0 = const()[name = tensor<string, []>("op_2672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2672_cast_fp16 = einsum(equation = var_2672_equation_0, values = (var_2604_cast_fp16_1, var_2654_cast_fp16))[name = tensor<string, []>("op_2672_cast_fp16")];
            tensor<string, []> var_2674_equation_0 = const()[name = tensor<string, []>("op_2674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2674_cast_fp16 = einsum(equation = var_2674_equation_0, values = (var_2604_cast_fp16_2, var_2655_cast_fp16))[name = tensor<string, []>("op_2674_cast_fp16")];
            tensor<string, []> var_2676_equation_0 = const()[name = tensor<string, []>("op_2676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2676_cast_fp16 = einsum(equation = var_2676_equation_0, values = (var_2604_cast_fp16_3, var_2656_cast_fp16))[name = tensor<string, []>("op_2676_cast_fp16")];
            tensor<string, []> var_2678_equation_0 = const()[name = tensor<string, []>("op_2678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2678_cast_fp16 = einsum(equation = var_2678_equation_0, values = (var_2604_cast_fp16_4, var_2657_cast_fp16))[name = tensor<string, []>("op_2678_cast_fp16")];
            tensor<string, []> var_2680_equation_0 = const()[name = tensor<string, []>("op_2680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2680_cast_fp16 = einsum(equation = var_2680_equation_0, values = (var_2604_cast_fp16_5, var_2658_cast_fp16))[name = tensor<string, []>("op_2680_cast_fp16")];
            tensor<string, []> var_2682_equation_0 = const()[name = tensor<string, []>("op_2682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2682_cast_fp16 = einsum(equation = var_2682_equation_0, values = (var_2604_cast_fp16_6, var_2659_cast_fp16))[name = tensor<string, []>("op_2682_cast_fp16")];
            tensor<string, []> var_2684_equation_0 = const()[name = tensor<string, []>("op_2684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2684_cast_fp16 = einsum(equation = var_2684_equation_0, values = (var_2604_cast_fp16_7, var_2660_cast_fp16))[name = tensor<string, []>("op_2684_cast_fp16")];
            tensor<string, []> var_2686_equation_0 = const()[name = tensor<string, []>("op_2686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2686_cast_fp16 = einsum(equation = var_2686_equation_0, values = (var_2604_cast_fp16_8, var_2661_cast_fp16))[name = tensor<string, []>("op_2686_cast_fp16")];
            tensor<string, []> var_2688_equation_0 = const()[name = tensor<string, []>("op_2688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2688_cast_fp16 = einsum(equation = var_2688_equation_0, values = (var_2604_cast_fp16_9, var_2662_cast_fp16))[name = tensor<string, []>("op_2688_cast_fp16")];
            tensor<string, []> var_2690_equation_0 = const()[name = tensor<string, []>("op_2690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2690_cast_fp16 = einsum(equation = var_2690_equation_0, values = (var_2604_cast_fp16_10, var_2663_cast_fp16))[name = tensor<string, []>("op_2690_cast_fp16")];
            tensor<string, []> var_2692_equation_0 = const()[name = tensor<string, []>("op_2692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2692_cast_fp16 = einsum(equation = var_2692_equation_0, values = (var_2604_cast_fp16_11, var_2664_cast_fp16))[name = tensor<string, []>("op_2692_cast_fp16")];
            tensor<string, []> var_2694_equation_0 = const()[name = tensor<string, []>("op_2694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2694_cast_fp16 = einsum(equation = var_2694_equation_0, values = (var_2604_cast_fp16_12, var_2665_cast_fp16))[name = tensor<string, []>("op_2694_cast_fp16")];
            tensor<string, []> var_2696_equation_0 = const()[name = tensor<string, []>("op_2696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2696_cast_fp16 = einsum(equation = var_2696_equation_0, values = (var_2604_cast_fp16_13, var_2666_cast_fp16))[name = tensor<string, []>("op_2696_cast_fp16")];
            tensor<string, []> var_2698_equation_0 = const()[name = tensor<string, []>("op_2698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2698_cast_fp16 = einsum(equation = var_2698_equation_0, values = (var_2604_cast_fp16_14, var_2667_cast_fp16))[name = tensor<string, []>("op_2698_cast_fp16")];
            tensor<string, []> var_2700_equation_0 = const()[name = tensor<string, []>("op_2700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2700_cast_fp16 = einsum(equation = var_2700_equation_0, values = (var_2604_cast_fp16_15, var_2668_cast_fp16))[name = tensor<string, []>("op_2700_cast_fp16")];
            tensor<bool, []> input_107_interleave_0 = const()[name = tensor<string, []>("input_107_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_107_cast_fp16 = concat(axis = var_2517, interleave = input_107_interleave_0, values = (var_2670_cast_fp16, var_2672_cast_fp16, var_2674_cast_fp16, var_2676_cast_fp16, var_2678_cast_fp16, var_2680_cast_fp16, var_2682_cast_fp16, var_2684_cast_fp16, var_2686_cast_fp16, var_2688_cast_fp16, var_2690_cast_fp16, var_2692_cast_fp16, var_2694_cast_fp16, var_2696_cast_fp16, var_2698_cast_fp16, var_2700_cast_fp16))[name = tensor<string, []>("input_107_cast_fp16")];
            tensor<string, []> var_2709_pad_type_0 = const()[name = tensor<string, []>("op_2709_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2709_strides_0 = const()[name = tensor<string, []>("op_2709_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2709_pad_0 = const()[name = tensor<string, []>("op_2709_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2709_dilations_0 = const()[name = tensor<string, []>("op_2709_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2709_groups_0 = const()[name = tensor<string, []>("op_2709_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_10_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66767296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67291648))), name = tensor<string, []>("blocks_10_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_10_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67291776)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2709_cast_fp16 = conv(bias = blocks_10_attn_out_bias_to_fp16, dilations = var_2709_dilations_0, groups = var_2709_groups_0, pad = var_2709_pad_0, pad_type = var_2709_pad_type_0, strides = var_2709_strides_0, weight = blocks_10_attn_out_weight_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor<string, []>("op_2709_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = var_2709_cast_fp16)[name = tensor<string, []>("inputs_43_cast_fp16")];
            tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_109_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_109_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67293888)))];
            tensor<fp16, [1024]> input_109_beta_0_to_fp16 = const()[name = tensor<string, []>("input_109_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67296000)))];
            tensor<fp16, []> var_2719_to_fp16 = const()[name = tensor<string, []>("op_2719_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = input_109_beta_0_to_fp16, epsilon = var_2719_to_fp16, gamma = input_109_gamma_0_to_fp16, x = inputs_43_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<string, []> input_111_pad_type_0 = const()[name = tensor<string, []>("input_111_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_111_strides_0 = const()[name = tensor<string, []>("input_111_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_111_pad_0 = const()[name = tensor<string, []>("input_111_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_111_dilations_0 = const()[name = tensor<string, []>("input_111_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_111_groups_0 = const()[name = tensor<string, []>("input_111_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_10_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67298112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69395328))), name = tensor<string, []>("blocks_10_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_10_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69395456)))];
            tensor<fp16, [1, 4096, 1, 600]> input_111_cast_fp16 = conv(bias = blocks_10_mlp_0_bias_to_fp16, dilations = input_111_dilations_0, groups = input_111_groups_0, pad = input_111_pad_0, pad_type = input_111_pad_type_0, strides = input_111_strides_0, weight = blocks_10_mlp_0_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
            tensor<string, []> input_113_mode_0 = const()[name = tensor<string, []>("input_113_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_113_cast_fp16 = gelu(mode = input_113_mode_0, x = input_111_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")];
            tensor<string, []> var_2745_pad_type_0 = const()[name = tensor<string, []>("op_2745_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2745_strides_0 = const()[name = tensor<string, []>("op_2745_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2745_pad_0 = const()[name = tensor<string, []>("op_2745_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2745_dilations_0 = const()[name = tensor<string, []>("op_2745_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2745_groups_0 = const()[name = tensor<string, []>("op_2745_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_10_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69403712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71500928))), name = tensor<string, []>("blocks_10_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_10_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71501056)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2745_cast_fp16 = conv(bias = blocks_10_mlp_2_bias_to_fp16, dilations = var_2745_dilations_0, groups = var_2745_groups_0, pad = var_2745_pad_0, pad_type = var_2745_pad_type_0, strides = var_2745_strides_0, weight = blocks_10_mlp_2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor<string, []>("op_2745_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = var_2745_cast_fp16)[name = tensor<string, []>("inputs_45_cast_fp16")];
            tensor<int32, []> var_2754 = const()[name = tensor<string, []>("op_2754"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_115_axes_0 = const()[name = tensor<string, []>("input_115_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_115_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_115_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71503168)))];
            tensor<fp16, [1024]> input_115_beta_0_to_fp16 = const()[name = tensor<string, []>("input_115_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71505280)))];
            tensor<fp16, []> var_2770_to_fp16 = const()[name = tensor<string, []>("op_2770_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_115_cast_fp16 = layer_norm(axes = input_115_axes_0, beta = input_115_beta_0_to_fp16, epsilon = var_2770_to_fp16, gamma = input_115_gamma_0_to_fp16, x = inputs_45_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<string, []> q_23_pad_type_0 = const()[name = tensor<string, []>("q_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_23_strides_0 = const()[name = tensor<string, []>("q_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_23_pad_0 = const()[name = tensor<string, []>("q_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_23_dilations_0 = const()[name = tensor<string, []>("q_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_23_groups_0 = const()[name = tensor<string, []>("q_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_2805_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71507392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72031744))), name = tensor<string, []>("op_2805_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_2805_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2805_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72031872)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2805_cast_fp16 = conv(bias = var_2805_bias_0_to_fp16, dilations = q_23_dilations_0, groups = q_23_groups_0, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = q_23_strides_0, weight = op_2805_weight_0_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor<string, []>("op_2805_cast_fp16")];
            tensor<string, []> k_23_pad_type_0 = const()[name = tensor<string, []>("k_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_23_strides_0 = const()[name = tensor<string, []>("k_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_23_pad_0 = const()[name = tensor<string, []>("k_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_23_dilations_0 = const()[name = tensor<string, []>("k_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_23_groups_0 = const()[name = tensor<string, []>("k_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_11_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72033984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72558336))), name = tensor<string, []>("blocks_11_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_23_cast_fp16 = conv(dilations = k_23_dilations_0, groups = k_23_groups_0, pad = k_23_pad_0, pad_type = k_23_pad_type_0, strides = k_23_strides_0, weight = blocks_11_attn_key_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor<string, []>("k_23_cast_fp16")];
            tensor<string, []> var_2803_pad_type_0 = const()[name = tensor<string, []>("op_2803_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2803_strides_0 = const()[name = tensor<string, []>("op_2803_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2803_pad_0 = const()[name = tensor<string, []>("op_2803_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2803_dilations_0 = const()[name = tensor<string, []>("op_2803_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2803_groups_0 = const()[name = tensor<string, []>("op_2803_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_11_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72558464))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73082816))), name = tensor<string, []>("blocks_11_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_11_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73082944)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2803_cast_fp16 = conv(bias = blocks_11_attn_value_bias_to_fp16, dilations = var_2803_dilations_0, groups = var_2803_groups_0, pad = var_2803_pad_0, pad_type = var_2803_pad_type_0, strides = var_2803_strides_0, weight = blocks_11_attn_value_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor<string, []>("op_2803_cast_fp16")];
            tensor<int32, [16]> tile_33 = const()[name = tensor<string, []>("tile_33"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2806_axis_0 = const()[name = tensor<string, []>("op_2806_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2806_cast_fp16_15 = split(axis = var_2806_axis_0, split_sizes = tile_33, x = var_2805_cast_fp16)[name = tensor<string, []>("op_2806_cast_fp16")];
            tensor<int32, [4]> var_2823_perm_0 = const()[name = tensor<string, []>("op_2823_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_34 = const()[name = tensor<string, []>("tile_34"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2824_axis_0 = const()[name = tensor<string, []>("op_2824_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_2823_cast_fp16 = transpose(perm = var_2823_perm_0, x = k_23_cast_fp16)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_2824_cast_fp16_15 = split(axis = var_2824_axis_0, split_sizes = tile_34, x = var_2823_cast_fp16)[name = tensor<string, []>("op_2824_cast_fp16")];
            tensor<int32, [16]> tile_35 = const()[name = tensor<string, []>("tile_35"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_2841_axis_0 = const()[name = tensor<string, []>("op_2841_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_2841_cast_fp16_15 = split(axis = var_2841_axis_0, split_sizes = tile_35, x = var_2803_cast_fp16)[name = tensor<string, []>("op_2841_cast_fp16")];
            tensor<string, []> aw_353_equation_0 = const()[name = tensor<string, []>("aw_353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_353_cast_fp16 = einsum(equation = aw_353_equation_0, values = (var_2824_cast_fp16_0, var_2806_cast_fp16_0))[name = tensor<string, []>("aw_353_cast_fp16")];
            tensor<string, []> aw_355_equation_0 = const()[name = tensor<string, []>("aw_355_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_355_cast_fp16 = einsum(equation = aw_355_equation_0, values = (var_2824_cast_fp16_1, var_2806_cast_fp16_1))[name = tensor<string, []>("aw_355_cast_fp16")];
            tensor<string, []> aw_357_equation_0 = const()[name = tensor<string, []>("aw_357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_357_cast_fp16 = einsum(equation = aw_357_equation_0, values = (var_2824_cast_fp16_2, var_2806_cast_fp16_2))[name = tensor<string, []>("aw_357_cast_fp16")];
            tensor<string, []> aw_359_equation_0 = const()[name = tensor<string, []>("aw_359_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_359_cast_fp16 = einsum(equation = aw_359_equation_0, values = (var_2824_cast_fp16_3, var_2806_cast_fp16_3))[name = tensor<string, []>("aw_359_cast_fp16")];
            tensor<string, []> aw_361_equation_0 = const()[name = tensor<string, []>("aw_361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_361_cast_fp16 = einsum(equation = aw_361_equation_0, values = (var_2824_cast_fp16_4, var_2806_cast_fp16_4))[name = tensor<string, []>("aw_361_cast_fp16")];
            tensor<string, []> aw_363_equation_0 = const()[name = tensor<string, []>("aw_363_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_363_cast_fp16 = einsum(equation = aw_363_equation_0, values = (var_2824_cast_fp16_5, var_2806_cast_fp16_5))[name = tensor<string, []>("aw_363_cast_fp16")];
            tensor<string, []> aw_365_equation_0 = const()[name = tensor<string, []>("aw_365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_365_cast_fp16 = einsum(equation = aw_365_equation_0, values = (var_2824_cast_fp16_6, var_2806_cast_fp16_6))[name = tensor<string, []>("aw_365_cast_fp16")];
            tensor<string, []> aw_367_equation_0 = const()[name = tensor<string, []>("aw_367_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_367_cast_fp16 = einsum(equation = aw_367_equation_0, values = (var_2824_cast_fp16_7, var_2806_cast_fp16_7))[name = tensor<string, []>("aw_367_cast_fp16")];
            tensor<string, []> aw_369_equation_0 = const()[name = tensor<string, []>("aw_369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_369_cast_fp16 = einsum(equation = aw_369_equation_0, values = (var_2824_cast_fp16_8, var_2806_cast_fp16_8))[name = tensor<string, []>("aw_369_cast_fp16")];
            tensor<string, []> aw_371_equation_0 = const()[name = tensor<string, []>("aw_371_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_371_cast_fp16 = einsum(equation = aw_371_equation_0, values = (var_2824_cast_fp16_9, var_2806_cast_fp16_9))[name = tensor<string, []>("aw_371_cast_fp16")];
            tensor<string, []> aw_373_equation_0 = const()[name = tensor<string, []>("aw_373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_373_cast_fp16 = einsum(equation = aw_373_equation_0, values = (var_2824_cast_fp16_10, var_2806_cast_fp16_10))[name = tensor<string, []>("aw_373_cast_fp16")];
            tensor<string, []> aw_375_equation_0 = const()[name = tensor<string, []>("aw_375_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_375_cast_fp16 = einsum(equation = aw_375_equation_0, values = (var_2824_cast_fp16_11, var_2806_cast_fp16_11))[name = tensor<string, []>("aw_375_cast_fp16")];
            tensor<string, []> aw_377_equation_0 = const()[name = tensor<string, []>("aw_377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_377_cast_fp16 = einsum(equation = aw_377_equation_0, values = (var_2824_cast_fp16_12, var_2806_cast_fp16_12))[name = tensor<string, []>("aw_377_cast_fp16")];
            tensor<string, []> aw_379_equation_0 = const()[name = tensor<string, []>("aw_379_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_379_cast_fp16 = einsum(equation = aw_379_equation_0, values = (var_2824_cast_fp16_13, var_2806_cast_fp16_13))[name = tensor<string, []>("aw_379_cast_fp16")];
            tensor<string, []> aw_381_equation_0 = const()[name = tensor<string, []>("aw_381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_381_cast_fp16 = einsum(equation = aw_381_equation_0, values = (var_2824_cast_fp16_14, var_2806_cast_fp16_14))[name = tensor<string, []>("aw_381_cast_fp16")];
            tensor<string, []> aw_383_equation_0 = const()[name = tensor<string, []>("aw_383_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_383_cast_fp16 = einsum(equation = aw_383_equation_0, values = (var_2824_cast_fp16_15, var_2806_cast_fp16_15))[name = tensor<string, []>("aw_383_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2890_cast_fp16 = softmax(axis = var_2754, x = aw_353_cast_fp16)[name = tensor<string, []>("op_2890_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2891_cast_fp16 = softmax(axis = var_2754, x = aw_355_cast_fp16)[name = tensor<string, []>("op_2891_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2892_cast_fp16 = softmax(axis = var_2754, x = aw_357_cast_fp16)[name = tensor<string, []>("op_2892_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2893_cast_fp16 = softmax(axis = var_2754, x = aw_359_cast_fp16)[name = tensor<string, []>("op_2893_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2894_cast_fp16 = softmax(axis = var_2754, x = aw_361_cast_fp16)[name = tensor<string, []>("op_2894_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2895_cast_fp16 = softmax(axis = var_2754, x = aw_363_cast_fp16)[name = tensor<string, []>("op_2895_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2896_cast_fp16 = softmax(axis = var_2754, x = aw_365_cast_fp16)[name = tensor<string, []>("op_2896_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2897_cast_fp16 = softmax(axis = var_2754, x = aw_367_cast_fp16)[name = tensor<string, []>("op_2897_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2898_cast_fp16 = softmax(axis = var_2754, x = aw_369_cast_fp16)[name = tensor<string, []>("op_2898_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2899_cast_fp16 = softmax(axis = var_2754, x = aw_371_cast_fp16)[name = tensor<string, []>("op_2899_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2900_cast_fp16 = softmax(axis = var_2754, x = aw_373_cast_fp16)[name = tensor<string, []>("op_2900_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2901_cast_fp16 = softmax(axis = var_2754, x = aw_375_cast_fp16)[name = tensor<string, []>("op_2901_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2902_cast_fp16 = softmax(axis = var_2754, x = aw_377_cast_fp16)[name = tensor<string, []>("op_2902_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2903_cast_fp16 = softmax(axis = var_2754, x = aw_379_cast_fp16)[name = tensor<string, []>("op_2903_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2904_cast_fp16 = softmax(axis = var_2754, x = aw_381_cast_fp16)[name = tensor<string, []>("op_2904_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_2905_cast_fp16 = softmax(axis = var_2754, x = aw_383_cast_fp16)[name = tensor<string, []>("op_2905_cast_fp16")];
            tensor<string, []> var_2907_equation_0 = const()[name = tensor<string, []>("op_2907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2841_cast_fp16_0, var_2890_cast_fp16))[name = tensor<string, []>("op_2907_cast_fp16")];
            tensor<string, []> var_2909_equation_0 = const()[name = tensor<string, []>("op_2909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2841_cast_fp16_1, var_2891_cast_fp16))[name = tensor<string, []>("op_2909_cast_fp16")];
            tensor<string, []> var_2911_equation_0 = const()[name = tensor<string, []>("op_2911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2841_cast_fp16_2, var_2892_cast_fp16))[name = tensor<string, []>("op_2911_cast_fp16")];
            tensor<string, []> var_2913_equation_0 = const()[name = tensor<string, []>("op_2913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2841_cast_fp16_3, var_2893_cast_fp16))[name = tensor<string, []>("op_2913_cast_fp16")];
            tensor<string, []> var_2915_equation_0 = const()[name = tensor<string, []>("op_2915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2841_cast_fp16_4, var_2894_cast_fp16))[name = tensor<string, []>("op_2915_cast_fp16")];
            tensor<string, []> var_2917_equation_0 = const()[name = tensor<string, []>("op_2917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2841_cast_fp16_5, var_2895_cast_fp16))[name = tensor<string, []>("op_2917_cast_fp16")];
            tensor<string, []> var_2919_equation_0 = const()[name = tensor<string, []>("op_2919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2841_cast_fp16_6, var_2896_cast_fp16))[name = tensor<string, []>("op_2919_cast_fp16")];
            tensor<string, []> var_2921_equation_0 = const()[name = tensor<string, []>("op_2921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2841_cast_fp16_7, var_2897_cast_fp16))[name = tensor<string, []>("op_2921_cast_fp16")];
            tensor<string, []> var_2923_equation_0 = const()[name = tensor<string, []>("op_2923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2841_cast_fp16_8, var_2898_cast_fp16))[name = tensor<string, []>("op_2923_cast_fp16")];
            tensor<string, []> var_2925_equation_0 = const()[name = tensor<string, []>("op_2925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2841_cast_fp16_9, var_2899_cast_fp16))[name = tensor<string, []>("op_2925_cast_fp16")];
            tensor<string, []> var_2927_equation_0 = const()[name = tensor<string, []>("op_2927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2841_cast_fp16_10, var_2900_cast_fp16))[name = tensor<string, []>("op_2927_cast_fp16")];
            tensor<string, []> var_2929_equation_0 = const()[name = tensor<string, []>("op_2929_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2841_cast_fp16_11, var_2901_cast_fp16))[name = tensor<string, []>("op_2929_cast_fp16")];
            tensor<string, []> var_2931_equation_0 = const()[name = tensor<string, []>("op_2931_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2841_cast_fp16_12, var_2902_cast_fp16))[name = tensor<string, []>("op_2931_cast_fp16")];
            tensor<string, []> var_2933_equation_0 = const()[name = tensor<string, []>("op_2933_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2841_cast_fp16_13, var_2903_cast_fp16))[name = tensor<string, []>("op_2933_cast_fp16")];
            tensor<string, []> var_2935_equation_0 = const()[name = tensor<string, []>("op_2935_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2841_cast_fp16_14, var_2904_cast_fp16))[name = tensor<string, []>("op_2935_cast_fp16")];
            tensor<string, []> var_2937_equation_0 = const()[name = tensor<string, []>("op_2937_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2841_cast_fp16_15, var_2905_cast_fp16))[name = tensor<string, []>("op_2937_cast_fp16")];
            tensor<bool, []> input_117_interleave_0 = const()[name = tensor<string, []>("input_117_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_117_cast_fp16 = concat(axis = var_2754, interleave = input_117_interleave_0, values = (var_2907_cast_fp16, var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16, var_2917_cast_fp16, var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16, var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16, var_2933_cast_fp16, var_2935_cast_fp16, var_2937_cast_fp16))[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<string, []> var_2946_pad_type_0 = const()[name = tensor<string, []>("op_2946_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2946_strides_0 = const()[name = tensor<string, []>("op_2946_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2946_pad_0 = const()[name = tensor<string, []>("op_2946_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2946_dilations_0 = const()[name = tensor<string, []>("op_2946_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2946_groups_0 = const()[name = tensor<string, []>("op_2946_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_11_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73085056))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73609408))), name = tensor<string, []>("blocks_11_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_11_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73609536)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2946_cast_fp16 = conv(bias = blocks_11_attn_out_bias_to_fp16, dilations = var_2946_dilations_0, groups = var_2946_groups_0, pad = var_2946_pad_0, pad_type = var_2946_pad_type_0, strides = var_2946_strides_0, weight = blocks_11_attn_out_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = tensor<string, []>("op_2946_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = var_2946_cast_fp16)[name = tensor<string, []>("inputs_47_cast_fp16")];
            tensor<int32, [1]> input_119_axes_0 = const()[name = tensor<string, []>("input_119_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_119_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_119_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73611648)))];
            tensor<fp16, [1024]> input_119_beta_0_to_fp16 = const()[name = tensor<string, []>("input_119_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73613760)))];
            tensor<fp16, []> var_2956_to_fp16 = const()[name = tensor<string, []>("op_2956_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_119_cast_fp16 = layer_norm(axes = input_119_axes_0, beta = input_119_beta_0_to_fp16, epsilon = var_2956_to_fp16, gamma = input_119_gamma_0_to_fp16, x = inputs_47_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")];
            tensor<string, []> input_121_pad_type_0 = const()[name = tensor<string, []>("input_121_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_121_strides_0 = const()[name = tensor<string, []>("input_121_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_121_pad_0 = const()[name = tensor<string, []>("input_121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_121_dilations_0 = const()[name = tensor<string, []>("input_121_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_121_groups_0 = const()[name = tensor<string, []>("input_121_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_11_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73615872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75713088))), name = tensor<string, []>("blocks_11_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_11_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75713216)))];
            tensor<fp16, [1, 4096, 1, 600]> input_121_cast_fp16 = conv(bias = blocks_11_mlp_0_bias_to_fp16, dilations = input_121_dilations_0, groups = input_121_groups_0, pad = input_121_pad_0, pad_type = input_121_pad_type_0, strides = input_121_strides_0, weight = blocks_11_mlp_0_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<string, []> input_123_mode_0 = const()[name = tensor<string, []>("input_123_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_123_cast_fp16 = gelu(mode = input_123_mode_0, x = input_121_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<string, []> var_2982_pad_type_0 = const()[name = tensor<string, []>("op_2982_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_2982_strides_0 = const()[name = tensor<string, []>("op_2982_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2982_pad_0 = const()[name = tensor<string, []>("op_2982_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2982_dilations_0 = const()[name = tensor<string, []>("op_2982_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_2982_groups_0 = const()[name = tensor<string, []>("op_2982_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_11_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75721472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77818688))), name = tensor<string, []>("blocks_11_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_11_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77818816)))];
            tensor<fp16, [1, 1024, 1, 600]> var_2982_cast_fp16 = conv(bias = blocks_11_mlp_2_bias_to_fp16, dilations = var_2982_dilations_0, groups = var_2982_groups_0, pad = var_2982_pad_0, pad_type = var_2982_pad_type_0, strides = var_2982_strides_0, weight = blocks_11_mlp_2_weight_to_fp16_palettized, x = input_123_cast_fp16)[name = tensor<string, []>("op_2982_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = var_2982_cast_fp16)[name = tensor<string, []>("inputs_49_cast_fp16")];
            tensor<int32, []> var_2991 = const()[name = tensor<string, []>("op_2991"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_125_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_125_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77820928)))];
            tensor<fp16, [1024]> input_125_beta_0_to_fp16 = const()[name = tensor<string, []>("input_125_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77823040)))];
            tensor<fp16, []> var_3007_to_fp16 = const()[name = tensor<string, []>("op_3007_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = input_125_beta_0_to_fp16, epsilon = var_3007_to_fp16, gamma = input_125_gamma_0_to_fp16, x = inputs_49_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<string, []> q_25_pad_type_0 = const()[name = tensor<string, []>("q_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_25_strides_0 = const()[name = tensor<string, []>("q_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_25_pad_0 = const()[name = tensor<string, []>("q_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_25_dilations_0 = const()[name = tensor<string, []>("q_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_25_groups_0 = const()[name = tensor<string, []>("q_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_3042_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77825152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78349504))), name = tensor<string, []>("op_3042_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_3042_bias_0_to_fp16 = const()[name = tensor<string, []>("op_3042_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78349632)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3042_cast_fp16 = conv(bias = var_3042_bias_0_to_fp16, dilations = q_25_dilations_0, groups = q_25_groups_0, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = q_25_strides_0, weight = op_3042_weight_0_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("op_3042_cast_fp16")];
            tensor<string, []> k_25_pad_type_0 = const()[name = tensor<string, []>("k_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_25_strides_0 = const()[name = tensor<string, []>("k_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_25_pad_0 = const()[name = tensor<string, []>("k_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_25_dilations_0 = const()[name = tensor<string, []>("k_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_25_groups_0 = const()[name = tensor<string, []>("k_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_12_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78351744))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78876096))), name = tensor<string, []>("blocks_12_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = blocks_12_attn_key_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("k_25_cast_fp16")];
            tensor<string, []> var_3040_pad_type_0 = const()[name = tensor<string, []>("op_3040_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3040_strides_0 = const()[name = tensor<string, []>("op_3040_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3040_pad_0 = const()[name = tensor<string, []>("op_3040_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3040_dilations_0 = const()[name = tensor<string, []>("op_3040_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3040_groups_0 = const()[name = tensor<string, []>("op_3040_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_12_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78876224))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79400576))), name = tensor<string, []>("blocks_12_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_12_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79400704)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3040_cast_fp16 = conv(bias = blocks_12_attn_value_bias_to_fp16, dilations = var_3040_dilations_0, groups = var_3040_groups_0, pad = var_3040_pad_0, pad_type = var_3040_pad_type_0, strides = var_3040_strides_0, weight = blocks_12_attn_value_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("op_3040_cast_fp16")];
            tensor<int32, [16]> tile_36 = const()[name = tensor<string, []>("tile_36"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3043_axis_0 = const()[name = tensor<string, []>("op_3043_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3043_cast_fp16_15 = split(axis = var_3043_axis_0, split_sizes = tile_36, x = var_3042_cast_fp16)[name = tensor<string, []>("op_3043_cast_fp16")];
            tensor<int32, [4]> var_3060_perm_0 = const()[name = tensor<string, []>("op_3060_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_37 = const()[name = tensor<string, []>("tile_37"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3061_axis_0 = const()[name = tensor<string, []>("op_3061_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_3060_cast_fp16 = transpose(perm = var_3060_perm_0, x = k_25_cast_fp16)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_3061_cast_fp16_15 = split(axis = var_3061_axis_0, split_sizes = tile_37, x = var_3060_cast_fp16)[name = tensor<string, []>("op_3061_cast_fp16")];
            tensor<int32, [16]> tile_38 = const()[name = tensor<string, []>("tile_38"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3078_axis_0 = const()[name = tensor<string, []>("op_3078_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3078_cast_fp16_15 = split(axis = var_3078_axis_0, split_sizes = tile_38, x = var_3040_cast_fp16)[name = tensor<string, []>("op_3078_cast_fp16")];
            tensor<string, []> aw_385_equation_0 = const()[name = tensor<string, []>("aw_385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_385_cast_fp16 = einsum(equation = aw_385_equation_0, values = (var_3061_cast_fp16_0, var_3043_cast_fp16_0))[name = tensor<string, []>("aw_385_cast_fp16")];
            tensor<string, []> aw_387_equation_0 = const()[name = tensor<string, []>("aw_387_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_387_cast_fp16 = einsum(equation = aw_387_equation_0, values = (var_3061_cast_fp16_1, var_3043_cast_fp16_1))[name = tensor<string, []>("aw_387_cast_fp16")];
            tensor<string, []> aw_389_equation_0 = const()[name = tensor<string, []>("aw_389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_389_cast_fp16 = einsum(equation = aw_389_equation_0, values = (var_3061_cast_fp16_2, var_3043_cast_fp16_2))[name = tensor<string, []>("aw_389_cast_fp16")];
            tensor<string, []> aw_391_equation_0 = const()[name = tensor<string, []>("aw_391_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_391_cast_fp16 = einsum(equation = aw_391_equation_0, values = (var_3061_cast_fp16_3, var_3043_cast_fp16_3))[name = tensor<string, []>("aw_391_cast_fp16")];
            tensor<string, []> aw_393_equation_0 = const()[name = tensor<string, []>("aw_393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_393_cast_fp16 = einsum(equation = aw_393_equation_0, values = (var_3061_cast_fp16_4, var_3043_cast_fp16_4))[name = tensor<string, []>("aw_393_cast_fp16")];
            tensor<string, []> aw_395_equation_0 = const()[name = tensor<string, []>("aw_395_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_395_cast_fp16 = einsum(equation = aw_395_equation_0, values = (var_3061_cast_fp16_5, var_3043_cast_fp16_5))[name = tensor<string, []>("aw_395_cast_fp16")];
            tensor<string, []> aw_397_equation_0 = const()[name = tensor<string, []>("aw_397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_397_cast_fp16 = einsum(equation = aw_397_equation_0, values = (var_3061_cast_fp16_6, var_3043_cast_fp16_6))[name = tensor<string, []>("aw_397_cast_fp16")];
            tensor<string, []> aw_399_equation_0 = const()[name = tensor<string, []>("aw_399_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_399_cast_fp16 = einsum(equation = aw_399_equation_0, values = (var_3061_cast_fp16_7, var_3043_cast_fp16_7))[name = tensor<string, []>("aw_399_cast_fp16")];
            tensor<string, []> aw_401_equation_0 = const()[name = tensor<string, []>("aw_401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_401_cast_fp16 = einsum(equation = aw_401_equation_0, values = (var_3061_cast_fp16_8, var_3043_cast_fp16_8))[name = tensor<string, []>("aw_401_cast_fp16")];
            tensor<string, []> aw_403_equation_0 = const()[name = tensor<string, []>("aw_403_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_403_cast_fp16 = einsum(equation = aw_403_equation_0, values = (var_3061_cast_fp16_9, var_3043_cast_fp16_9))[name = tensor<string, []>("aw_403_cast_fp16")];
            tensor<string, []> aw_405_equation_0 = const()[name = tensor<string, []>("aw_405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_405_cast_fp16 = einsum(equation = aw_405_equation_0, values = (var_3061_cast_fp16_10, var_3043_cast_fp16_10))[name = tensor<string, []>("aw_405_cast_fp16")];
            tensor<string, []> aw_407_equation_0 = const()[name = tensor<string, []>("aw_407_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_407_cast_fp16 = einsum(equation = aw_407_equation_0, values = (var_3061_cast_fp16_11, var_3043_cast_fp16_11))[name = tensor<string, []>("aw_407_cast_fp16")];
            tensor<string, []> aw_409_equation_0 = const()[name = tensor<string, []>("aw_409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_409_cast_fp16 = einsum(equation = aw_409_equation_0, values = (var_3061_cast_fp16_12, var_3043_cast_fp16_12))[name = tensor<string, []>("aw_409_cast_fp16")];
            tensor<string, []> aw_411_equation_0 = const()[name = tensor<string, []>("aw_411_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_411_cast_fp16 = einsum(equation = aw_411_equation_0, values = (var_3061_cast_fp16_13, var_3043_cast_fp16_13))[name = tensor<string, []>("aw_411_cast_fp16")];
            tensor<string, []> aw_413_equation_0 = const()[name = tensor<string, []>("aw_413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_413_cast_fp16 = einsum(equation = aw_413_equation_0, values = (var_3061_cast_fp16_14, var_3043_cast_fp16_14))[name = tensor<string, []>("aw_413_cast_fp16")];
            tensor<string, []> aw_415_equation_0 = const()[name = tensor<string, []>("aw_415_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_415_cast_fp16 = einsum(equation = aw_415_equation_0, values = (var_3061_cast_fp16_15, var_3043_cast_fp16_15))[name = tensor<string, []>("aw_415_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3127_cast_fp16 = softmax(axis = var_2991, x = aw_385_cast_fp16)[name = tensor<string, []>("op_3127_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3128_cast_fp16 = softmax(axis = var_2991, x = aw_387_cast_fp16)[name = tensor<string, []>("op_3128_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3129_cast_fp16 = softmax(axis = var_2991, x = aw_389_cast_fp16)[name = tensor<string, []>("op_3129_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3130_cast_fp16 = softmax(axis = var_2991, x = aw_391_cast_fp16)[name = tensor<string, []>("op_3130_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3131_cast_fp16 = softmax(axis = var_2991, x = aw_393_cast_fp16)[name = tensor<string, []>("op_3131_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3132_cast_fp16 = softmax(axis = var_2991, x = aw_395_cast_fp16)[name = tensor<string, []>("op_3132_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3133_cast_fp16 = softmax(axis = var_2991, x = aw_397_cast_fp16)[name = tensor<string, []>("op_3133_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3134_cast_fp16 = softmax(axis = var_2991, x = aw_399_cast_fp16)[name = tensor<string, []>("op_3134_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3135_cast_fp16 = softmax(axis = var_2991, x = aw_401_cast_fp16)[name = tensor<string, []>("op_3135_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3136_cast_fp16 = softmax(axis = var_2991, x = aw_403_cast_fp16)[name = tensor<string, []>("op_3136_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3137_cast_fp16 = softmax(axis = var_2991, x = aw_405_cast_fp16)[name = tensor<string, []>("op_3137_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3138_cast_fp16 = softmax(axis = var_2991, x = aw_407_cast_fp16)[name = tensor<string, []>("op_3138_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3139_cast_fp16 = softmax(axis = var_2991, x = aw_409_cast_fp16)[name = tensor<string, []>("op_3139_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3140_cast_fp16 = softmax(axis = var_2991, x = aw_411_cast_fp16)[name = tensor<string, []>("op_3140_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3141_cast_fp16 = softmax(axis = var_2991, x = aw_413_cast_fp16)[name = tensor<string, []>("op_3141_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3142_cast_fp16 = softmax(axis = var_2991, x = aw_415_cast_fp16)[name = tensor<string, []>("op_3142_cast_fp16")];
            tensor<string, []> var_3144_equation_0 = const()[name = tensor<string, []>("op_3144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3144_cast_fp16 = einsum(equation = var_3144_equation_0, values = (var_3078_cast_fp16_0, var_3127_cast_fp16))[name = tensor<string, []>("op_3144_cast_fp16")];
            tensor<string, []> var_3146_equation_0 = const()[name = tensor<string, []>("op_3146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3146_cast_fp16 = einsum(equation = var_3146_equation_0, values = (var_3078_cast_fp16_1, var_3128_cast_fp16))[name = tensor<string, []>("op_3146_cast_fp16")];
            tensor<string, []> var_3148_equation_0 = const()[name = tensor<string, []>("op_3148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3148_cast_fp16 = einsum(equation = var_3148_equation_0, values = (var_3078_cast_fp16_2, var_3129_cast_fp16))[name = tensor<string, []>("op_3148_cast_fp16")];
            tensor<string, []> var_3150_equation_0 = const()[name = tensor<string, []>("op_3150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3150_cast_fp16 = einsum(equation = var_3150_equation_0, values = (var_3078_cast_fp16_3, var_3130_cast_fp16))[name = tensor<string, []>("op_3150_cast_fp16")];
            tensor<string, []> var_3152_equation_0 = const()[name = tensor<string, []>("op_3152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3152_cast_fp16 = einsum(equation = var_3152_equation_0, values = (var_3078_cast_fp16_4, var_3131_cast_fp16))[name = tensor<string, []>("op_3152_cast_fp16")];
            tensor<string, []> var_3154_equation_0 = const()[name = tensor<string, []>("op_3154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3154_cast_fp16 = einsum(equation = var_3154_equation_0, values = (var_3078_cast_fp16_5, var_3132_cast_fp16))[name = tensor<string, []>("op_3154_cast_fp16")];
            tensor<string, []> var_3156_equation_0 = const()[name = tensor<string, []>("op_3156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3156_cast_fp16 = einsum(equation = var_3156_equation_0, values = (var_3078_cast_fp16_6, var_3133_cast_fp16))[name = tensor<string, []>("op_3156_cast_fp16")];
            tensor<string, []> var_3158_equation_0 = const()[name = tensor<string, []>("op_3158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3158_cast_fp16 = einsum(equation = var_3158_equation_0, values = (var_3078_cast_fp16_7, var_3134_cast_fp16))[name = tensor<string, []>("op_3158_cast_fp16")];
            tensor<string, []> var_3160_equation_0 = const()[name = tensor<string, []>("op_3160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3160_cast_fp16 = einsum(equation = var_3160_equation_0, values = (var_3078_cast_fp16_8, var_3135_cast_fp16))[name = tensor<string, []>("op_3160_cast_fp16")];
            tensor<string, []> var_3162_equation_0 = const()[name = tensor<string, []>("op_3162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3162_cast_fp16 = einsum(equation = var_3162_equation_0, values = (var_3078_cast_fp16_9, var_3136_cast_fp16))[name = tensor<string, []>("op_3162_cast_fp16")];
            tensor<string, []> var_3164_equation_0 = const()[name = tensor<string, []>("op_3164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3164_cast_fp16 = einsum(equation = var_3164_equation_0, values = (var_3078_cast_fp16_10, var_3137_cast_fp16))[name = tensor<string, []>("op_3164_cast_fp16")];
            tensor<string, []> var_3166_equation_0 = const()[name = tensor<string, []>("op_3166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3166_cast_fp16 = einsum(equation = var_3166_equation_0, values = (var_3078_cast_fp16_11, var_3138_cast_fp16))[name = tensor<string, []>("op_3166_cast_fp16")];
            tensor<string, []> var_3168_equation_0 = const()[name = tensor<string, []>("op_3168_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3168_cast_fp16 = einsum(equation = var_3168_equation_0, values = (var_3078_cast_fp16_12, var_3139_cast_fp16))[name = tensor<string, []>("op_3168_cast_fp16")];
            tensor<string, []> var_3170_equation_0 = const()[name = tensor<string, []>("op_3170_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3170_cast_fp16 = einsum(equation = var_3170_equation_0, values = (var_3078_cast_fp16_13, var_3140_cast_fp16))[name = tensor<string, []>("op_3170_cast_fp16")];
            tensor<string, []> var_3172_equation_0 = const()[name = tensor<string, []>("op_3172_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3172_cast_fp16 = einsum(equation = var_3172_equation_0, values = (var_3078_cast_fp16_14, var_3141_cast_fp16))[name = tensor<string, []>("op_3172_cast_fp16")];
            tensor<string, []> var_3174_equation_0 = const()[name = tensor<string, []>("op_3174_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3174_cast_fp16 = einsum(equation = var_3174_equation_0, values = (var_3078_cast_fp16_15, var_3142_cast_fp16))[name = tensor<string, []>("op_3174_cast_fp16")];
            tensor<bool, []> input_127_interleave_0 = const()[name = tensor<string, []>("input_127_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_127_cast_fp16 = concat(axis = var_2991, interleave = input_127_interleave_0, values = (var_3144_cast_fp16, var_3146_cast_fp16, var_3148_cast_fp16, var_3150_cast_fp16, var_3152_cast_fp16, var_3154_cast_fp16, var_3156_cast_fp16, var_3158_cast_fp16, var_3160_cast_fp16, var_3162_cast_fp16, var_3164_cast_fp16, var_3166_cast_fp16, var_3168_cast_fp16, var_3170_cast_fp16, var_3172_cast_fp16, var_3174_cast_fp16))[name = tensor<string, []>("input_127_cast_fp16")];
            tensor<string, []> var_3183_pad_type_0 = const()[name = tensor<string, []>("op_3183_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3183_strides_0 = const()[name = tensor<string, []>("op_3183_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3183_pad_0 = const()[name = tensor<string, []>("op_3183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3183_dilations_0 = const()[name = tensor<string, []>("op_3183_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3183_groups_0 = const()[name = tensor<string, []>("op_3183_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_12_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79402816))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79927168))), name = tensor<string, []>("blocks_12_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_12_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79927296)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3183_cast_fp16 = conv(bias = blocks_12_attn_out_bias_to_fp16, dilations = var_3183_dilations_0, groups = var_3183_groups_0, pad = var_3183_pad_0, pad_type = var_3183_pad_type_0, strides = var_3183_strides_0, weight = blocks_12_attn_out_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor<string, []>("op_3183_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = var_3183_cast_fp16)[name = tensor<string, []>("inputs_51_cast_fp16")];
            tensor<int32, [1]> input_129_axes_0 = const()[name = tensor<string, []>("input_129_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_129_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_129_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79929408)))];
            tensor<fp16, [1024]> input_129_beta_0_to_fp16 = const()[name = tensor<string, []>("input_129_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79931520)))];
            tensor<fp16, []> var_3193_to_fp16 = const()[name = tensor<string, []>("op_3193_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_129_cast_fp16 = layer_norm(axes = input_129_axes_0, beta = input_129_beta_0_to_fp16, epsilon = var_3193_to_fp16, gamma = input_129_gamma_0_to_fp16, x = inputs_51_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<string, []> input_131_pad_type_0 = const()[name = tensor<string, []>("input_131_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_131_strides_0 = const()[name = tensor<string, []>("input_131_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_131_pad_0 = const()[name = tensor<string, []>("input_131_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_131_dilations_0 = const()[name = tensor<string, []>("input_131_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_131_groups_0 = const()[name = tensor<string, []>("input_131_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_12_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79933632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82030848))), name = tensor<string, []>("blocks_12_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_12_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82030976)))];
            tensor<fp16, [1, 4096, 1, 600]> input_131_cast_fp16 = conv(bias = blocks_12_mlp_0_bias_to_fp16, dilations = input_131_dilations_0, groups = input_131_groups_0, pad = input_131_pad_0, pad_type = input_131_pad_type_0, strides = input_131_strides_0, weight = blocks_12_mlp_0_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")];
            tensor<string, []> input_133_mode_0 = const()[name = tensor<string, []>("input_133_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = input_131_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<string, []> var_3219_pad_type_0 = const()[name = tensor<string, []>("op_3219_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3219_strides_0 = const()[name = tensor<string, []>("op_3219_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3219_pad_0 = const()[name = tensor<string, []>("op_3219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3219_dilations_0 = const()[name = tensor<string, []>("op_3219_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3219_groups_0 = const()[name = tensor<string, []>("op_3219_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_12_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82039232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84136448))), name = tensor<string, []>("blocks_12_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_12_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84136576)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3219_cast_fp16 = conv(bias = blocks_12_mlp_2_bias_to_fp16, dilations = var_3219_dilations_0, groups = var_3219_groups_0, pad = var_3219_pad_0, pad_type = var_3219_pad_type_0, strides = var_3219_strides_0, weight = blocks_12_mlp_2_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = tensor<string, []>("op_3219_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = var_3219_cast_fp16)[name = tensor<string, []>("inputs_53_cast_fp16")];
            tensor<int32, []> var_3228 = const()[name = tensor<string, []>("op_3228"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_135_axes_0 = const()[name = tensor<string, []>("input_135_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_135_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_135_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84138688)))];
            tensor<fp16, [1024]> input_135_beta_0_to_fp16 = const()[name = tensor<string, []>("input_135_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84140800)))];
            tensor<fp16, []> var_3244_to_fp16 = const()[name = tensor<string, []>("op_3244_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_135_cast_fp16 = layer_norm(axes = input_135_axes_0, beta = input_135_beta_0_to_fp16, epsilon = var_3244_to_fp16, gamma = input_135_gamma_0_to_fp16, x = inputs_53_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<string, []> q_27_pad_type_0 = const()[name = tensor<string, []>("q_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_27_strides_0 = const()[name = tensor<string, []>("q_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_27_pad_0 = const()[name = tensor<string, []>("q_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_27_dilations_0 = const()[name = tensor<string, []>("q_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_27_groups_0 = const()[name = tensor<string, []>("q_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_3279_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84142912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84667264))), name = tensor<string, []>("op_3279_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_3279_bias_0_to_fp16 = const()[name = tensor<string, []>("op_3279_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84667392)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3279_cast_fp16 = conv(bias = var_3279_bias_0_to_fp16, dilations = q_27_dilations_0, groups = q_27_groups_0, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = q_27_strides_0, weight = op_3279_weight_0_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor<string, []>("op_3279_cast_fp16")];
            tensor<string, []> k_27_pad_type_0 = const()[name = tensor<string, []>("k_27_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_27_strides_0 = const()[name = tensor<string, []>("k_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_27_pad_0 = const()[name = tensor<string, []>("k_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_27_dilations_0 = const()[name = tensor<string, []>("k_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_27_groups_0 = const()[name = tensor<string, []>("k_27_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_13_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84669504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85193856))), name = tensor<string, []>("blocks_13_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_27_cast_fp16 = conv(dilations = k_27_dilations_0, groups = k_27_groups_0, pad = k_27_pad_0, pad_type = k_27_pad_type_0, strides = k_27_strides_0, weight = blocks_13_attn_key_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor<string, []>("k_27_cast_fp16")];
            tensor<string, []> var_3277_pad_type_0 = const()[name = tensor<string, []>("op_3277_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3277_strides_0 = const()[name = tensor<string, []>("op_3277_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3277_pad_0 = const()[name = tensor<string, []>("op_3277_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3277_dilations_0 = const()[name = tensor<string, []>("op_3277_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3277_groups_0 = const()[name = tensor<string, []>("op_3277_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_13_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85193984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85718336))), name = tensor<string, []>("blocks_13_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_13_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85718464)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3277_cast_fp16 = conv(bias = blocks_13_attn_value_bias_to_fp16, dilations = var_3277_dilations_0, groups = var_3277_groups_0, pad = var_3277_pad_0, pad_type = var_3277_pad_type_0, strides = var_3277_strides_0, weight = blocks_13_attn_value_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor<string, []>("op_3277_cast_fp16")];
            tensor<int32, [16]> tile_39 = const()[name = tensor<string, []>("tile_39"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3280_axis_0 = const()[name = tensor<string, []>("op_3280_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3280_cast_fp16_15 = split(axis = var_3280_axis_0, split_sizes = tile_39, x = var_3279_cast_fp16)[name = tensor<string, []>("op_3280_cast_fp16")];
            tensor<int32, [4]> var_3297_perm_0 = const()[name = tensor<string, []>("op_3297_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_40 = const()[name = tensor<string, []>("tile_40"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3298_axis_0 = const()[name = tensor<string, []>("op_3298_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_3297_cast_fp16 = transpose(perm = var_3297_perm_0, x = k_27_cast_fp16)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_3298_cast_fp16_15 = split(axis = var_3298_axis_0, split_sizes = tile_40, x = var_3297_cast_fp16)[name = tensor<string, []>("op_3298_cast_fp16")];
            tensor<int32, [16]> tile_41 = const()[name = tensor<string, []>("tile_41"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3315_axis_0 = const()[name = tensor<string, []>("op_3315_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3315_cast_fp16_15 = split(axis = var_3315_axis_0, split_sizes = tile_41, x = var_3277_cast_fp16)[name = tensor<string, []>("op_3315_cast_fp16")];
            tensor<string, []> aw_417_equation_0 = const()[name = tensor<string, []>("aw_417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_417_cast_fp16 = einsum(equation = aw_417_equation_0, values = (var_3298_cast_fp16_0, var_3280_cast_fp16_0))[name = tensor<string, []>("aw_417_cast_fp16")];
            tensor<string, []> aw_419_equation_0 = const()[name = tensor<string, []>("aw_419_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_419_cast_fp16 = einsum(equation = aw_419_equation_0, values = (var_3298_cast_fp16_1, var_3280_cast_fp16_1))[name = tensor<string, []>("aw_419_cast_fp16")];
            tensor<string, []> aw_421_equation_0 = const()[name = tensor<string, []>("aw_421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_421_cast_fp16 = einsum(equation = aw_421_equation_0, values = (var_3298_cast_fp16_2, var_3280_cast_fp16_2))[name = tensor<string, []>("aw_421_cast_fp16")];
            tensor<string, []> aw_423_equation_0 = const()[name = tensor<string, []>("aw_423_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_423_cast_fp16 = einsum(equation = aw_423_equation_0, values = (var_3298_cast_fp16_3, var_3280_cast_fp16_3))[name = tensor<string, []>("aw_423_cast_fp16")];
            tensor<string, []> aw_425_equation_0 = const()[name = tensor<string, []>("aw_425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_425_cast_fp16 = einsum(equation = aw_425_equation_0, values = (var_3298_cast_fp16_4, var_3280_cast_fp16_4))[name = tensor<string, []>("aw_425_cast_fp16")];
            tensor<string, []> aw_427_equation_0 = const()[name = tensor<string, []>("aw_427_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_427_cast_fp16 = einsum(equation = aw_427_equation_0, values = (var_3298_cast_fp16_5, var_3280_cast_fp16_5))[name = tensor<string, []>("aw_427_cast_fp16")];
            tensor<string, []> aw_429_equation_0 = const()[name = tensor<string, []>("aw_429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_429_cast_fp16 = einsum(equation = aw_429_equation_0, values = (var_3298_cast_fp16_6, var_3280_cast_fp16_6))[name = tensor<string, []>("aw_429_cast_fp16")];
            tensor<string, []> aw_431_equation_0 = const()[name = tensor<string, []>("aw_431_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_431_cast_fp16 = einsum(equation = aw_431_equation_0, values = (var_3298_cast_fp16_7, var_3280_cast_fp16_7))[name = tensor<string, []>("aw_431_cast_fp16")];
            tensor<string, []> aw_433_equation_0 = const()[name = tensor<string, []>("aw_433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_433_cast_fp16 = einsum(equation = aw_433_equation_0, values = (var_3298_cast_fp16_8, var_3280_cast_fp16_8))[name = tensor<string, []>("aw_433_cast_fp16")];
            tensor<string, []> aw_435_equation_0 = const()[name = tensor<string, []>("aw_435_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_435_cast_fp16 = einsum(equation = aw_435_equation_0, values = (var_3298_cast_fp16_9, var_3280_cast_fp16_9))[name = tensor<string, []>("aw_435_cast_fp16")];
            tensor<string, []> aw_437_equation_0 = const()[name = tensor<string, []>("aw_437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_437_cast_fp16 = einsum(equation = aw_437_equation_0, values = (var_3298_cast_fp16_10, var_3280_cast_fp16_10))[name = tensor<string, []>("aw_437_cast_fp16")];
            tensor<string, []> aw_439_equation_0 = const()[name = tensor<string, []>("aw_439_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_439_cast_fp16 = einsum(equation = aw_439_equation_0, values = (var_3298_cast_fp16_11, var_3280_cast_fp16_11))[name = tensor<string, []>("aw_439_cast_fp16")];
            tensor<string, []> aw_441_equation_0 = const()[name = tensor<string, []>("aw_441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_441_cast_fp16 = einsum(equation = aw_441_equation_0, values = (var_3298_cast_fp16_12, var_3280_cast_fp16_12))[name = tensor<string, []>("aw_441_cast_fp16")];
            tensor<string, []> aw_443_equation_0 = const()[name = tensor<string, []>("aw_443_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_443_cast_fp16 = einsum(equation = aw_443_equation_0, values = (var_3298_cast_fp16_13, var_3280_cast_fp16_13))[name = tensor<string, []>("aw_443_cast_fp16")];
            tensor<string, []> aw_445_equation_0 = const()[name = tensor<string, []>("aw_445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_445_cast_fp16 = einsum(equation = aw_445_equation_0, values = (var_3298_cast_fp16_14, var_3280_cast_fp16_14))[name = tensor<string, []>("aw_445_cast_fp16")];
            tensor<string, []> aw_447_equation_0 = const()[name = tensor<string, []>("aw_447_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_447_cast_fp16 = einsum(equation = aw_447_equation_0, values = (var_3298_cast_fp16_15, var_3280_cast_fp16_15))[name = tensor<string, []>("aw_447_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3364_cast_fp16 = softmax(axis = var_3228, x = aw_417_cast_fp16)[name = tensor<string, []>("op_3364_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3365_cast_fp16 = softmax(axis = var_3228, x = aw_419_cast_fp16)[name = tensor<string, []>("op_3365_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3366_cast_fp16 = softmax(axis = var_3228, x = aw_421_cast_fp16)[name = tensor<string, []>("op_3366_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3367_cast_fp16 = softmax(axis = var_3228, x = aw_423_cast_fp16)[name = tensor<string, []>("op_3367_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3368_cast_fp16 = softmax(axis = var_3228, x = aw_425_cast_fp16)[name = tensor<string, []>("op_3368_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3369_cast_fp16 = softmax(axis = var_3228, x = aw_427_cast_fp16)[name = tensor<string, []>("op_3369_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3370_cast_fp16 = softmax(axis = var_3228, x = aw_429_cast_fp16)[name = tensor<string, []>("op_3370_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3371_cast_fp16 = softmax(axis = var_3228, x = aw_431_cast_fp16)[name = tensor<string, []>("op_3371_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3372_cast_fp16 = softmax(axis = var_3228, x = aw_433_cast_fp16)[name = tensor<string, []>("op_3372_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3373_cast_fp16 = softmax(axis = var_3228, x = aw_435_cast_fp16)[name = tensor<string, []>("op_3373_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3374_cast_fp16 = softmax(axis = var_3228, x = aw_437_cast_fp16)[name = tensor<string, []>("op_3374_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3375_cast_fp16 = softmax(axis = var_3228, x = aw_439_cast_fp16)[name = tensor<string, []>("op_3375_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3376_cast_fp16 = softmax(axis = var_3228, x = aw_441_cast_fp16)[name = tensor<string, []>("op_3376_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3377_cast_fp16 = softmax(axis = var_3228, x = aw_443_cast_fp16)[name = tensor<string, []>("op_3377_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3378_cast_fp16 = softmax(axis = var_3228, x = aw_445_cast_fp16)[name = tensor<string, []>("op_3378_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3379_cast_fp16 = softmax(axis = var_3228, x = aw_447_cast_fp16)[name = tensor<string, []>("op_3379_cast_fp16")];
            tensor<string, []> var_3381_equation_0 = const()[name = tensor<string, []>("op_3381_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3381_cast_fp16 = einsum(equation = var_3381_equation_0, values = (var_3315_cast_fp16_0, var_3364_cast_fp16))[name = tensor<string, []>("op_3381_cast_fp16")];
            tensor<string, []> var_3383_equation_0 = const()[name = tensor<string, []>("op_3383_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3383_cast_fp16 = einsum(equation = var_3383_equation_0, values = (var_3315_cast_fp16_1, var_3365_cast_fp16))[name = tensor<string, []>("op_3383_cast_fp16")];
            tensor<string, []> var_3385_equation_0 = const()[name = tensor<string, []>("op_3385_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3385_cast_fp16 = einsum(equation = var_3385_equation_0, values = (var_3315_cast_fp16_2, var_3366_cast_fp16))[name = tensor<string, []>("op_3385_cast_fp16")];
            tensor<string, []> var_3387_equation_0 = const()[name = tensor<string, []>("op_3387_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3387_cast_fp16 = einsum(equation = var_3387_equation_0, values = (var_3315_cast_fp16_3, var_3367_cast_fp16))[name = tensor<string, []>("op_3387_cast_fp16")];
            tensor<string, []> var_3389_equation_0 = const()[name = tensor<string, []>("op_3389_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3389_cast_fp16 = einsum(equation = var_3389_equation_0, values = (var_3315_cast_fp16_4, var_3368_cast_fp16))[name = tensor<string, []>("op_3389_cast_fp16")];
            tensor<string, []> var_3391_equation_0 = const()[name = tensor<string, []>("op_3391_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3391_cast_fp16 = einsum(equation = var_3391_equation_0, values = (var_3315_cast_fp16_5, var_3369_cast_fp16))[name = tensor<string, []>("op_3391_cast_fp16")];
            tensor<string, []> var_3393_equation_0 = const()[name = tensor<string, []>("op_3393_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3393_cast_fp16 = einsum(equation = var_3393_equation_0, values = (var_3315_cast_fp16_6, var_3370_cast_fp16))[name = tensor<string, []>("op_3393_cast_fp16")];
            tensor<string, []> var_3395_equation_0 = const()[name = tensor<string, []>("op_3395_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3395_cast_fp16 = einsum(equation = var_3395_equation_0, values = (var_3315_cast_fp16_7, var_3371_cast_fp16))[name = tensor<string, []>("op_3395_cast_fp16")];
            tensor<string, []> var_3397_equation_0 = const()[name = tensor<string, []>("op_3397_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3397_cast_fp16 = einsum(equation = var_3397_equation_0, values = (var_3315_cast_fp16_8, var_3372_cast_fp16))[name = tensor<string, []>("op_3397_cast_fp16")];
            tensor<string, []> var_3399_equation_0 = const()[name = tensor<string, []>("op_3399_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3399_cast_fp16 = einsum(equation = var_3399_equation_0, values = (var_3315_cast_fp16_9, var_3373_cast_fp16))[name = tensor<string, []>("op_3399_cast_fp16")];
            tensor<string, []> var_3401_equation_0 = const()[name = tensor<string, []>("op_3401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3401_cast_fp16 = einsum(equation = var_3401_equation_0, values = (var_3315_cast_fp16_10, var_3374_cast_fp16))[name = tensor<string, []>("op_3401_cast_fp16")];
            tensor<string, []> var_3403_equation_0 = const()[name = tensor<string, []>("op_3403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3403_cast_fp16 = einsum(equation = var_3403_equation_0, values = (var_3315_cast_fp16_11, var_3375_cast_fp16))[name = tensor<string, []>("op_3403_cast_fp16")];
            tensor<string, []> var_3405_equation_0 = const()[name = tensor<string, []>("op_3405_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3405_cast_fp16 = einsum(equation = var_3405_equation_0, values = (var_3315_cast_fp16_12, var_3376_cast_fp16))[name = tensor<string, []>("op_3405_cast_fp16")];
            tensor<string, []> var_3407_equation_0 = const()[name = tensor<string, []>("op_3407_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3407_cast_fp16 = einsum(equation = var_3407_equation_0, values = (var_3315_cast_fp16_13, var_3377_cast_fp16))[name = tensor<string, []>("op_3407_cast_fp16")];
            tensor<string, []> var_3409_equation_0 = const()[name = tensor<string, []>("op_3409_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3409_cast_fp16 = einsum(equation = var_3409_equation_0, values = (var_3315_cast_fp16_14, var_3378_cast_fp16))[name = tensor<string, []>("op_3409_cast_fp16")];
            tensor<string, []> var_3411_equation_0 = const()[name = tensor<string, []>("op_3411_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3411_cast_fp16 = einsum(equation = var_3411_equation_0, values = (var_3315_cast_fp16_15, var_3379_cast_fp16))[name = tensor<string, []>("op_3411_cast_fp16")];
            tensor<bool, []> input_137_interleave_0 = const()[name = tensor<string, []>("input_137_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_137_cast_fp16 = concat(axis = var_3228, interleave = input_137_interleave_0, values = (var_3381_cast_fp16, var_3383_cast_fp16, var_3385_cast_fp16, var_3387_cast_fp16, var_3389_cast_fp16, var_3391_cast_fp16, var_3393_cast_fp16, var_3395_cast_fp16, var_3397_cast_fp16, var_3399_cast_fp16, var_3401_cast_fp16, var_3403_cast_fp16, var_3405_cast_fp16, var_3407_cast_fp16, var_3409_cast_fp16, var_3411_cast_fp16))[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<string, []> var_3420_pad_type_0 = const()[name = tensor<string, []>("op_3420_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3420_strides_0 = const()[name = tensor<string, []>("op_3420_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3420_pad_0 = const()[name = tensor<string, []>("op_3420_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3420_dilations_0 = const()[name = tensor<string, []>("op_3420_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3420_groups_0 = const()[name = tensor<string, []>("op_3420_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_13_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85720576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86244928))), name = tensor<string, []>("blocks_13_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_13_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86245056)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3420_cast_fp16 = conv(bias = blocks_13_attn_out_bias_to_fp16, dilations = var_3420_dilations_0, groups = var_3420_groups_0, pad = var_3420_pad_0, pad_type = var_3420_pad_type_0, strides = var_3420_strides_0, weight = blocks_13_attn_out_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor<string, []>("op_3420_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = var_3420_cast_fp16)[name = tensor<string, []>("inputs_55_cast_fp16")];
            tensor<int32, [1]> input_139_axes_0 = const()[name = tensor<string, []>("input_139_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_139_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_139_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86247168)))];
            tensor<fp16, [1024]> input_139_beta_0_to_fp16 = const()[name = tensor<string, []>("input_139_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86249280)))];
            tensor<fp16, []> var_3430_to_fp16 = const()[name = tensor<string, []>("op_3430_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_139_cast_fp16 = layer_norm(axes = input_139_axes_0, beta = input_139_beta_0_to_fp16, epsilon = var_3430_to_fp16, gamma = input_139_gamma_0_to_fp16, x = inputs_55_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<string, []> input_141_pad_type_0 = const()[name = tensor<string, []>("input_141_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_141_strides_0 = const()[name = tensor<string, []>("input_141_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_141_pad_0 = const()[name = tensor<string, []>("input_141_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_141_dilations_0 = const()[name = tensor<string, []>("input_141_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_141_groups_0 = const()[name = tensor<string, []>("input_141_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_13_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86251392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88348608))), name = tensor<string, []>("blocks_13_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_13_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88348736)))];
            tensor<fp16, [1, 4096, 1, 600]> input_141_cast_fp16 = conv(bias = blocks_13_mlp_0_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = blocks_13_mlp_0_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")];
            tensor<string, []> input_143_mode_0 = const()[name = tensor<string, []>("input_143_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<string, []> var_3456_pad_type_0 = const()[name = tensor<string, []>("op_3456_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3456_strides_0 = const()[name = tensor<string, []>("op_3456_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3456_pad_0 = const()[name = tensor<string, []>("op_3456_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3456_dilations_0 = const()[name = tensor<string, []>("op_3456_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3456_groups_0 = const()[name = tensor<string, []>("op_3456_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_13_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88356992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90454208))), name = tensor<string, []>("blocks_13_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_13_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90454336)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3456_cast_fp16 = conv(bias = blocks_13_mlp_2_bias_to_fp16, dilations = var_3456_dilations_0, groups = var_3456_groups_0, pad = var_3456_pad_0, pad_type = var_3456_pad_type_0, strides = var_3456_strides_0, weight = blocks_13_mlp_2_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = tensor<string, []>("op_3456_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = var_3456_cast_fp16)[name = tensor<string, []>("inputs_57_cast_fp16")];
            tensor<int32, []> var_3465 = const()[name = tensor<string, []>("op_3465"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_145_axes_0 = const()[name = tensor<string, []>("input_145_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_145_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_145_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90456448)))];
            tensor<fp16, [1024]> input_145_beta_0_to_fp16 = const()[name = tensor<string, []>("input_145_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90458560)))];
            tensor<fp16, []> var_3481_to_fp16 = const()[name = tensor<string, []>("op_3481_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_145_cast_fp16 = layer_norm(axes = input_145_axes_0, beta = input_145_beta_0_to_fp16, epsilon = var_3481_to_fp16, gamma = input_145_gamma_0_to_fp16, x = inputs_57_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<string, []> q_29_pad_type_0 = const()[name = tensor<string, []>("q_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_29_strides_0 = const()[name = tensor<string, []>("q_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_29_pad_0 = const()[name = tensor<string, []>("q_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_29_dilations_0 = const()[name = tensor<string, []>("q_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_29_groups_0 = const()[name = tensor<string, []>("q_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_3516_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90460672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90985024))), name = tensor<string, []>("op_3516_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_3516_bias_0_to_fp16 = const()[name = tensor<string, []>("op_3516_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90985152)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3516_cast_fp16 = conv(bias = var_3516_bias_0_to_fp16, dilations = q_29_dilations_0, groups = q_29_groups_0, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = q_29_strides_0, weight = op_3516_weight_0_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("op_3516_cast_fp16")];
            tensor<string, []> k_29_pad_type_0 = const()[name = tensor<string, []>("k_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_29_strides_0 = const()[name = tensor<string, []>("k_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_29_pad_0 = const()[name = tensor<string, []>("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_29_dilations_0 = const()[name = tensor<string, []>("k_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_29_groups_0 = const()[name = tensor<string, []>("k_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_14_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90987264))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91511616))), name = tensor<string, []>("blocks_14_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = blocks_14_attn_key_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("k_29_cast_fp16")];
            tensor<string, []> var_3514_pad_type_0 = const()[name = tensor<string, []>("op_3514_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3514_strides_0 = const()[name = tensor<string, []>("op_3514_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3514_pad_0 = const()[name = tensor<string, []>("op_3514_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3514_dilations_0 = const()[name = tensor<string, []>("op_3514_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3514_groups_0 = const()[name = tensor<string, []>("op_3514_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_14_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91511744))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92036096))), name = tensor<string, []>("blocks_14_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_14_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92036224)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3514_cast_fp16 = conv(bias = blocks_14_attn_value_bias_to_fp16, dilations = var_3514_dilations_0, groups = var_3514_groups_0, pad = var_3514_pad_0, pad_type = var_3514_pad_type_0, strides = var_3514_strides_0, weight = blocks_14_attn_value_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("op_3514_cast_fp16")];
            tensor<int32, [16]> tile_42 = const()[name = tensor<string, []>("tile_42"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3517_axis_0 = const()[name = tensor<string, []>("op_3517_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3517_cast_fp16_15 = split(axis = var_3517_axis_0, split_sizes = tile_42, x = var_3516_cast_fp16)[name = tensor<string, []>("op_3517_cast_fp16")];
            tensor<int32, [4]> var_3534_perm_0 = const()[name = tensor<string, []>("op_3534_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_43 = const()[name = tensor<string, []>("tile_43"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3535_axis_0 = const()[name = tensor<string, []>("op_3535_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_3534_cast_fp16 = transpose(perm = var_3534_perm_0, x = k_29_cast_fp16)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_3535_cast_fp16_15 = split(axis = var_3535_axis_0, split_sizes = tile_43, x = var_3534_cast_fp16)[name = tensor<string, []>("op_3535_cast_fp16")];
            tensor<int32, [16]> tile_44 = const()[name = tensor<string, []>("tile_44"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3552_axis_0 = const()[name = tensor<string, []>("op_3552_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3552_cast_fp16_15 = split(axis = var_3552_axis_0, split_sizes = tile_44, x = var_3514_cast_fp16)[name = tensor<string, []>("op_3552_cast_fp16")];
            tensor<string, []> aw_449_equation_0 = const()[name = tensor<string, []>("aw_449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_449_cast_fp16 = einsum(equation = aw_449_equation_0, values = (var_3535_cast_fp16_0, var_3517_cast_fp16_0))[name = tensor<string, []>("aw_449_cast_fp16")];
            tensor<string, []> aw_451_equation_0 = const()[name = tensor<string, []>("aw_451_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_451_cast_fp16 = einsum(equation = aw_451_equation_0, values = (var_3535_cast_fp16_1, var_3517_cast_fp16_1))[name = tensor<string, []>("aw_451_cast_fp16")];
            tensor<string, []> aw_453_equation_0 = const()[name = tensor<string, []>("aw_453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_453_cast_fp16 = einsum(equation = aw_453_equation_0, values = (var_3535_cast_fp16_2, var_3517_cast_fp16_2))[name = tensor<string, []>("aw_453_cast_fp16")];
            tensor<string, []> aw_455_equation_0 = const()[name = tensor<string, []>("aw_455_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_455_cast_fp16 = einsum(equation = aw_455_equation_0, values = (var_3535_cast_fp16_3, var_3517_cast_fp16_3))[name = tensor<string, []>("aw_455_cast_fp16")];
            tensor<string, []> aw_457_equation_0 = const()[name = tensor<string, []>("aw_457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_457_cast_fp16 = einsum(equation = aw_457_equation_0, values = (var_3535_cast_fp16_4, var_3517_cast_fp16_4))[name = tensor<string, []>("aw_457_cast_fp16")];
            tensor<string, []> aw_459_equation_0 = const()[name = tensor<string, []>("aw_459_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_459_cast_fp16 = einsum(equation = aw_459_equation_0, values = (var_3535_cast_fp16_5, var_3517_cast_fp16_5))[name = tensor<string, []>("aw_459_cast_fp16")];
            tensor<string, []> aw_461_equation_0 = const()[name = tensor<string, []>("aw_461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_461_cast_fp16 = einsum(equation = aw_461_equation_0, values = (var_3535_cast_fp16_6, var_3517_cast_fp16_6))[name = tensor<string, []>("aw_461_cast_fp16")];
            tensor<string, []> aw_463_equation_0 = const()[name = tensor<string, []>("aw_463_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_463_cast_fp16 = einsum(equation = aw_463_equation_0, values = (var_3535_cast_fp16_7, var_3517_cast_fp16_7))[name = tensor<string, []>("aw_463_cast_fp16")];
            tensor<string, []> aw_465_equation_0 = const()[name = tensor<string, []>("aw_465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_465_cast_fp16 = einsum(equation = aw_465_equation_0, values = (var_3535_cast_fp16_8, var_3517_cast_fp16_8))[name = tensor<string, []>("aw_465_cast_fp16")];
            tensor<string, []> aw_467_equation_0 = const()[name = tensor<string, []>("aw_467_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_467_cast_fp16 = einsum(equation = aw_467_equation_0, values = (var_3535_cast_fp16_9, var_3517_cast_fp16_9))[name = tensor<string, []>("aw_467_cast_fp16")];
            tensor<string, []> aw_469_equation_0 = const()[name = tensor<string, []>("aw_469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_469_cast_fp16 = einsum(equation = aw_469_equation_0, values = (var_3535_cast_fp16_10, var_3517_cast_fp16_10))[name = tensor<string, []>("aw_469_cast_fp16")];
            tensor<string, []> aw_471_equation_0 = const()[name = tensor<string, []>("aw_471_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_471_cast_fp16 = einsum(equation = aw_471_equation_0, values = (var_3535_cast_fp16_11, var_3517_cast_fp16_11))[name = tensor<string, []>("aw_471_cast_fp16")];
            tensor<string, []> aw_473_equation_0 = const()[name = tensor<string, []>("aw_473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_473_cast_fp16 = einsum(equation = aw_473_equation_0, values = (var_3535_cast_fp16_12, var_3517_cast_fp16_12))[name = tensor<string, []>("aw_473_cast_fp16")];
            tensor<string, []> aw_475_equation_0 = const()[name = tensor<string, []>("aw_475_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_475_cast_fp16 = einsum(equation = aw_475_equation_0, values = (var_3535_cast_fp16_13, var_3517_cast_fp16_13))[name = tensor<string, []>("aw_475_cast_fp16")];
            tensor<string, []> aw_477_equation_0 = const()[name = tensor<string, []>("aw_477_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_477_cast_fp16 = einsum(equation = aw_477_equation_0, values = (var_3535_cast_fp16_14, var_3517_cast_fp16_14))[name = tensor<string, []>("aw_477_cast_fp16")];
            tensor<string, []> aw_479_equation_0 = const()[name = tensor<string, []>("aw_479_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_479_cast_fp16 = einsum(equation = aw_479_equation_0, values = (var_3535_cast_fp16_15, var_3517_cast_fp16_15))[name = tensor<string, []>("aw_479_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3601_cast_fp16 = softmax(axis = var_3465, x = aw_449_cast_fp16)[name = tensor<string, []>("op_3601_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3602_cast_fp16 = softmax(axis = var_3465, x = aw_451_cast_fp16)[name = tensor<string, []>("op_3602_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3603_cast_fp16 = softmax(axis = var_3465, x = aw_453_cast_fp16)[name = tensor<string, []>("op_3603_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3604_cast_fp16 = softmax(axis = var_3465, x = aw_455_cast_fp16)[name = tensor<string, []>("op_3604_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3605_cast_fp16 = softmax(axis = var_3465, x = aw_457_cast_fp16)[name = tensor<string, []>("op_3605_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3606_cast_fp16 = softmax(axis = var_3465, x = aw_459_cast_fp16)[name = tensor<string, []>("op_3606_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3607_cast_fp16 = softmax(axis = var_3465, x = aw_461_cast_fp16)[name = tensor<string, []>("op_3607_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3608_cast_fp16 = softmax(axis = var_3465, x = aw_463_cast_fp16)[name = tensor<string, []>("op_3608_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3609_cast_fp16 = softmax(axis = var_3465, x = aw_465_cast_fp16)[name = tensor<string, []>("op_3609_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3610_cast_fp16 = softmax(axis = var_3465, x = aw_467_cast_fp16)[name = tensor<string, []>("op_3610_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3611_cast_fp16 = softmax(axis = var_3465, x = aw_469_cast_fp16)[name = tensor<string, []>("op_3611_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3612_cast_fp16 = softmax(axis = var_3465, x = aw_471_cast_fp16)[name = tensor<string, []>("op_3612_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3613_cast_fp16 = softmax(axis = var_3465, x = aw_473_cast_fp16)[name = tensor<string, []>("op_3613_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3614_cast_fp16 = softmax(axis = var_3465, x = aw_475_cast_fp16)[name = tensor<string, []>("op_3614_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3615_cast_fp16 = softmax(axis = var_3465, x = aw_477_cast_fp16)[name = tensor<string, []>("op_3615_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3616_cast_fp16 = softmax(axis = var_3465, x = aw_479_cast_fp16)[name = tensor<string, []>("op_3616_cast_fp16")];
            tensor<string, []> var_3618_equation_0 = const()[name = tensor<string, []>("op_3618_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3618_cast_fp16 = einsum(equation = var_3618_equation_0, values = (var_3552_cast_fp16_0, var_3601_cast_fp16))[name = tensor<string, []>("op_3618_cast_fp16")];
            tensor<string, []> var_3620_equation_0 = const()[name = tensor<string, []>("op_3620_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3620_cast_fp16 = einsum(equation = var_3620_equation_0, values = (var_3552_cast_fp16_1, var_3602_cast_fp16))[name = tensor<string, []>("op_3620_cast_fp16")];
            tensor<string, []> var_3622_equation_0 = const()[name = tensor<string, []>("op_3622_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3622_cast_fp16 = einsum(equation = var_3622_equation_0, values = (var_3552_cast_fp16_2, var_3603_cast_fp16))[name = tensor<string, []>("op_3622_cast_fp16")];
            tensor<string, []> var_3624_equation_0 = const()[name = tensor<string, []>("op_3624_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3624_cast_fp16 = einsum(equation = var_3624_equation_0, values = (var_3552_cast_fp16_3, var_3604_cast_fp16))[name = tensor<string, []>("op_3624_cast_fp16")];
            tensor<string, []> var_3626_equation_0 = const()[name = tensor<string, []>("op_3626_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3626_cast_fp16 = einsum(equation = var_3626_equation_0, values = (var_3552_cast_fp16_4, var_3605_cast_fp16))[name = tensor<string, []>("op_3626_cast_fp16")];
            tensor<string, []> var_3628_equation_0 = const()[name = tensor<string, []>("op_3628_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3628_cast_fp16 = einsum(equation = var_3628_equation_0, values = (var_3552_cast_fp16_5, var_3606_cast_fp16))[name = tensor<string, []>("op_3628_cast_fp16")];
            tensor<string, []> var_3630_equation_0 = const()[name = tensor<string, []>("op_3630_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3630_cast_fp16 = einsum(equation = var_3630_equation_0, values = (var_3552_cast_fp16_6, var_3607_cast_fp16))[name = tensor<string, []>("op_3630_cast_fp16")];
            tensor<string, []> var_3632_equation_0 = const()[name = tensor<string, []>("op_3632_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3632_cast_fp16 = einsum(equation = var_3632_equation_0, values = (var_3552_cast_fp16_7, var_3608_cast_fp16))[name = tensor<string, []>("op_3632_cast_fp16")];
            tensor<string, []> var_3634_equation_0 = const()[name = tensor<string, []>("op_3634_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3634_cast_fp16 = einsum(equation = var_3634_equation_0, values = (var_3552_cast_fp16_8, var_3609_cast_fp16))[name = tensor<string, []>("op_3634_cast_fp16")];
            tensor<string, []> var_3636_equation_0 = const()[name = tensor<string, []>("op_3636_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3636_cast_fp16 = einsum(equation = var_3636_equation_0, values = (var_3552_cast_fp16_9, var_3610_cast_fp16))[name = tensor<string, []>("op_3636_cast_fp16")];
            tensor<string, []> var_3638_equation_0 = const()[name = tensor<string, []>("op_3638_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3638_cast_fp16 = einsum(equation = var_3638_equation_0, values = (var_3552_cast_fp16_10, var_3611_cast_fp16))[name = tensor<string, []>("op_3638_cast_fp16")];
            tensor<string, []> var_3640_equation_0 = const()[name = tensor<string, []>("op_3640_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3640_cast_fp16 = einsum(equation = var_3640_equation_0, values = (var_3552_cast_fp16_11, var_3612_cast_fp16))[name = tensor<string, []>("op_3640_cast_fp16")];
            tensor<string, []> var_3642_equation_0 = const()[name = tensor<string, []>("op_3642_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3642_cast_fp16 = einsum(equation = var_3642_equation_0, values = (var_3552_cast_fp16_12, var_3613_cast_fp16))[name = tensor<string, []>("op_3642_cast_fp16")];
            tensor<string, []> var_3644_equation_0 = const()[name = tensor<string, []>("op_3644_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3644_cast_fp16 = einsum(equation = var_3644_equation_0, values = (var_3552_cast_fp16_13, var_3614_cast_fp16))[name = tensor<string, []>("op_3644_cast_fp16")];
            tensor<string, []> var_3646_equation_0 = const()[name = tensor<string, []>("op_3646_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3646_cast_fp16 = einsum(equation = var_3646_equation_0, values = (var_3552_cast_fp16_14, var_3615_cast_fp16))[name = tensor<string, []>("op_3646_cast_fp16")];
            tensor<string, []> var_3648_equation_0 = const()[name = tensor<string, []>("op_3648_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3648_cast_fp16 = einsum(equation = var_3648_equation_0, values = (var_3552_cast_fp16_15, var_3616_cast_fp16))[name = tensor<string, []>("op_3648_cast_fp16")];
            tensor<bool, []> input_147_interleave_0 = const()[name = tensor<string, []>("input_147_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_147_cast_fp16 = concat(axis = var_3465, interleave = input_147_interleave_0, values = (var_3618_cast_fp16, var_3620_cast_fp16, var_3622_cast_fp16, var_3624_cast_fp16, var_3626_cast_fp16, var_3628_cast_fp16, var_3630_cast_fp16, var_3632_cast_fp16, var_3634_cast_fp16, var_3636_cast_fp16, var_3638_cast_fp16, var_3640_cast_fp16, var_3642_cast_fp16, var_3644_cast_fp16, var_3646_cast_fp16, var_3648_cast_fp16))[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<string, []> var_3657_pad_type_0 = const()[name = tensor<string, []>("op_3657_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3657_strides_0 = const()[name = tensor<string, []>("op_3657_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3657_pad_0 = const()[name = tensor<string, []>("op_3657_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3657_dilations_0 = const()[name = tensor<string, []>("op_3657_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3657_groups_0 = const()[name = tensor<string, []>("op_3657_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_14_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92038336))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92562688))), name = tensor<string, []>("blocks_14_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_14_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92562816)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3657_cast_fp16 = conv(bias = blocks_14_attn_out_bias_to_fp16, dilations = var_3657_dilations_0, groups = var_3657_groups_0, pad = var_3657_pad_0, pad_type = var_3657_pad_type_0, strides = var_3657_strides_0, weight = blocks_14_attn_out_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor<string, []>("op_3657_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = var_3657_cast_fp16)[name = tensor<string, []>("inputs_59_cast_fp16")];
            tensor<int32, [1]> input_149_axes_0 = const()[name = tensor<string, []>("input_149_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_149_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_149_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92564928)))];
            tensor<fp16, [1024]> input_149_beta_0_to_fp16 = const()[name = tensor<string, []>("input_149_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92567040)))];
            tensor<fp16, []> var_3667_to_fp16 = const()[name = tensor<string, []>("op_3667_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_149_cast_fp16 = layer_norm(axes = input_149_axes_0, beta = input_149_beta_0_to_fp16, epsilon = var_3667_to_fp16, gamma = input_149_gamma_0_to_fp16, x = inputs_59_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")];
            tensor<string, []> input_151_pad_type_0 = const()[name = tensor<string, []>("input_151_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_151_strides_0 = const()[name = tensor<string, []>("input_151_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_151_pad_0 = const()[name = tensor<string, []>("input_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_151_dilations_0 = const()[name = tensor<string, []>("input_151_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_151_groups_0 = const()[name = tensor<string, []>("input_151_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_14_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92569152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94666368))), name = tensor<string, []>("blocks_14_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_14_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94666496)))];
            tensor<fp16, [1, 4096, 1, 600]> input_151_cast_fp16 = conv(bias = blocks_14_mlp_0_bias_to_fp16, dilations = input_151_dilations_0, groups = input_151_groups_0, pad = input_151_pad_0, pad_type = input_151_pad_type_0, strides = input_151_strides_0, weight = blocks_14_mlp_0_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<string, []> input_153_mode_0 = const()[name = tensor<string, []>("input_153_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_153_cast_fp16 = gelu(mode = input_153_mode_0, x = input_151_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")];
            tensor<string, []> var_3693_pad_type_0 = const()[name = tensor<string, []>("op_3693_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3693_strides_0 = const()[name = tensor<string, []>("op_3693_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3693_pad_0 = const()[name = tensor<string, []>("op_3693_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3693_dilations_0 = const()[name = tensor<string, []>("op_3693_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3693_groups_0 = const()[name = tensor<string, []>("op_3693_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_14_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94674752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96771968))), name = tensor<string, []>("blocks_14_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_14_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96772096)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3693_cast_fp16 = conv(bias = blocks_14_mlp_2_bias_to_fp16, dilations = var_3693_dilations_0, groups = var_3693_groups_0, pad = var_3693_pad_0, pad_type = var_3693_pad_type_0, strides = var_3693_strides_0, weight = blocks_14_mlp_2_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor<string, []>("op_3693_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = var_3693_cast_fp16)[name = tensor<string, []>("inputs_61_cast_fp16")];
            tensor<int32, []> var_3702 = const()[name = tensor<string, []>("op_3702"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_155_axes_0 = const()[name = tensor<string, []>("input_155_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_155_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_155_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96774208)))];
            tensor<fp16, [1024]> input_155_beta_0_to_fp16 = const()[name = tensor<string, []>("input_155_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96776320)))];
            tensor<fp16, []> var_3718_to_fp16 = const()[name = tensor<string, []>("op_3718_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_155_cast_fp16 = layer_norm(axes = input_155_axes_0, beta = input_155_beta_0_to_fp16, epsilon = var_3718_to_fp16, gamma = input_155_gamma_0_to_fp16, x = inputs_61_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<string, []> q_31_pad_type_0 = const()[name = tensor<string, []>("q_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_31_strides_0 = const()[name = tensor<string, []>("q_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_31_pad_0 = const()[name = tensor<string, []>("q_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_31_dilations_0 = const()[name = tensor<string, []>("q_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_31_groups_0 = const()[name = tensor<string, []>("q_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_3753_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96778432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97302784))), name = tensor<string, []>("op_3753_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_3753_bias_0_to_fp16 = const()[name = tensor<string, []>("op_3753_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97302912)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3753_cast_fp16 = conv(bias = var_3753_bias_0_to_fp16, dilations = q_31_dilations_0, groups = q_31_groups_0, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = q_31_strides_0, weight = op_3753_weight_0_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor<string, []>("op_3753_cast_fp16")];
            tensor<string, []> k_31_pad_type_0 = const()[name = tensor<string, []>("k_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_31_strides_0 = const()[name = tensor<string, []>("k_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_31_pad_0 = const()[name = tensor<string, []>("k_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_31_dilations_0 = const()[name = tensor<string, []>("k_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_31_groups_0 = const()[name = tensor<string, []>("k_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_15_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97305024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97829376))), name = tensor<string, []>("blocks_15_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_31_cast_fp16 = conv(dilations = k_31_dilations_0, groups = k_31_groups_0, pad = k_31_pad_0, pad_type = k_31_pad_type_0, strides = k_31_strides_0, weight = blocks_15_attn_key_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor<string, []>("k_31_cast_fp16")];
            tensor<string, []> var_3751_pad_type_0 = const()[name = tensor<string, []>("op_3751_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3751_strides_0 = const()[name = tensor<string, []>("op_3751_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3751_pad_0 = const()[name = tensor<string, []>("op_3751_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3751_dilations_0 = const()[name = tensor<string, []>("op_3751_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3751_groups_0 = const()[name = tensor<string, []>("op_3751_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_15_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97829504))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98353856))), name = tensor<string, []>("blocks_15_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_15_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98353984)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3751_cast_fp16 = conv(bias = blocks_15_attn_value_bias_to_fp16, dilations = var_3751_dilations_0, groups = var_3751_groups_0, pad = var_3751_pad_0, pad_type = var_3751_pad_type_0, strides = var_3751_strides_0, weight = blocks_15_attn_value_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor<string, []>("op_3751_cast_fp16")];
            tensor<int32, [16]> tile_45 = const()[name = tensor<string, []>("tile_45"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3754_axis_0 = const()[name = tensor<string, []>("op_3754_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3754_cast_fp16_15 = split(axis = var_3754_axis_0, split_sizes = tile_45, x = var_3753_cast_fp16)[name = tensor<string, []>("op_3754_cast_fp16")];
            tensor<int32, [4]> var_3771_perm_0 = const()[name = tensor<string, []>("op_3771_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_46 = const()[name = tensor<string, []>("tile_46"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3772_axis_0 = const()[name = tensor<string, []>("op_3772_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_3771_cast_fp16 = transpose(perm = var_3771_perm_0, x = k_31_cast_fp16)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_3772_cast_fp16_15 = split(axis = var_3772_axis_0, split_sizes = tile_46, x = var_3771_cast_fp16)[name = tensor<string, []>("op_3772_cast_fp16")];
            tensor<int32, [16]> tile_47 = const()[name = tensor<string, []>("tile_47"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3789_axis_0 = const()[name = tensor<string, []>("op_3789_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3789_cast_fp16_15 = split(axis = var_3789_axis_0, split_sizes = tile_47, x = var_3751_cast_fp16)[name = tensor<string, []>("op_3789_cast_fp16")];
            tensor<string, []> aw_481_equation_0 = const()[name = tensor<string, []>("aw_481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_481_cast_fp16 = einsum(equation = aw_481_equation_0, values = (var_3772_cast_fp16_0, var_3754_cast_fp16_0))[name = tensor<string, []>("aw_481_cast_fp16")];
            tensor<string, []> aw_483_equation_0 = const()[name = tensor<string, []>("aw_483_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_483_cast_fp16 = einsum(equation = aw_483_equation_0, values = (var_3772_cast_fp16_1, var_3754_cast_fp16_1))[name = tensor<string, []>("aw_483_cast_fp16")];
            tensor<string, []> aw_485_equation_0 = const()[name = tensor<string, []>("aw_485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_485_cast_fp16 = einsum(equation = aw_485_equation_0, values = (var_3772_cast_fp16_2, var_3754_cast_fp16_2))[name = tensor<string, []>("aw_485_cast_fp16")];
            tensor<string, []> aw_487_equation_0 = const()[name = tensor<string, []>("aw_487_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_487_cast_fp16 = einsum(equation = aw_487_equation_0, values = (var_3772_cast_fp16_3, var_3754_cast_fp16_3))[name = tensor<string, []>("aw_487_cast_fp16")];
            tensor<string, []> aw_489_equation_0 = const()[name = tensor<string, []>("aw_489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_489_cast_fp16 = einsum(equation = aw_489_equation_0, values = (var_3772_cast_fp16_4, var_3754_cast_fp16_4))[name = tensor<string, []>("aw_489_cast_fp16")];
            tensor<string, []> aw_491_equation_0 = const()[name = tensor<string, []>("aw_491_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_491_cast_fp16 = einsum(equation = aw_491_equation_0, values = (var_3772_cast_fp16_5, var_3754_cast_fp16_5))[name = tensor<string, []>("aw_491_cast_fp16")];
            tensor<string, []> aw_493_equation_0 = const()[name = tensor<string, []>("aw_493_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_493_cast_fp16 = einsum(equation = aw_493_equation_0, values = (var_3772_cast_fp16_6, var_3754_cast_fp16_6))[name = tensor<string, []>("aw_493_cast_fp16")];
            tensor<string, []> aw_495_equation_0 = const()[name = tensor<string, []>("aw_495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_495_cast_fp16 = einsum(equation = aw_495_equation_0, values = (var_3772_cast_fp16_7, var_3754_cast_fp16_7))[name = tensor<string, []>("aw_495_cast_fp16")];
            tensor<string, []> aw_497_equation_0 = const()[name = tensor<string, []>("aw_497_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_497_cast_fp16 = einsum(equation = aw_497_equation_0, values = (var_3772_cast_fp16_8, var_3754_cast_fp16_8))[name = tensor<string, []>("aw_497_cast_fp16")];
            tensor<string, []> aw_499_equation_0 = const()[name = tensor<string, []>("aw_499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_499_cast_fp16 = einsum(equation = aw_499_equation_0, values = (var_3772_cast_fp16_9, var_3754_cast_fp16_9))[name = tensor<string, []>("aw_499_cast_fp16")];
            tensor<string, []> aw_501_equation_0 = const()[name = tensor<string, []>("aw_501_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_501_cast_fp16 = einsum(equation = aw_501_equation_0, values = (var_3772_cast_fp16_10, var_3754_cast_fp16_10))[name = tensor<string, []>("aw_501_cast_fp16")];
            tensor<string, []> aw_503_equation_0 = const()[name = tensor<string, []>("aw_503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_503_cast_fp16 = einsum(equation = aw_503_equation_0, values = (var_3772_cast_fp16_11, var_3754_cast_fp16_11))[name = tensor<string, []>("aw_503_cast_fp16")];
            tensor<string, []> aw_505_equation_0 = const()[name = tensor<string, []>("aw_505_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_505_cast_fp16 = einsum(equation = aw_505_equation_0, values = (var_3772_cast_fp16_12, var_3754_cast_fp16_12))[name = tensor<string, []>("aw_505_cast_fp16")];
            tensor<string, []> aw_507_equation_0 = const()[name = tensor<string, []>("aw_507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_507_cast_fp16 = einsum(equation = aw_507_equation_0, values = (var_3772_cast_fp16_13, var_3754_cast_fp16_13))[name = tensor<string, []>("aw_507_cast_fp16")];
            tensor<string, []> aw_509_equation_0 = const()[name = tensor<string, []>("aw_509_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_509_cast_fp16 = einsum(equation = aw_509_equation_0, values = (var_3772_cast_fp16_14, var_3754_cast_fp16_14))[name = tensor<string, []>("aw_509_cast_fp16")];
            tensor<string, []> aw_511_equation_0 = const()[name = tensor<string, []>("aw_511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_511_cast_fp16 = einsum(equation = aw_511_equation_0, values = (var_3772_cast_fp16_15, var_3754_cast_fp16_15))[name = tensor<string, []>("aw_511_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3838_cast_fp16 = softmax(axis = var_3702, x = aw_481_cast_fp16)[name = tensor<string, []>("op_3838_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3839_cast_fp16 = softmax(axis = var_3702, x = aw_483_cast_fp16)[name = tensor<string, []>("op_3839_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3840_cast_fp16 = softmax(axis = var_3702, x = aw_485_cast_fp16)[name = tensor<string, []>("op_3840_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3841_cast_fp16 = softmax(axis = var_3702, x = aw_487_cast_fp16)[name = tensor<string, []>("op_3841_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3842_cast_fp16 = softmax(axis = var_3702, x = aw_489_cast_fp16)[name = tensor<string, []>("op_3842_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3843_cast_fp16 = softmax(axis = var_3702, x = aw_491_cast_fp16)[name = tensor<string, []>("op_3843_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3844_cast_fp16 = softmax(axis = var_3702, x = aw_493_cast_fp16)[name = tensor<string, []>("op_3844_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3845_cast_fp16 = softmax(axis = var_3702, x = aw_495_cast_fp16)[name = tensor<string, []>("op_3845_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3846_cast_fp16 = softmax(axis = var_3702, x = aw_497_cast_fp16)[name = tensor<string, []>("op_3846_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3847_cast_fp16 = softmax(axis = var_3702, x = aw_499_cast_fp16)[name = tensor<string, []>("op_3847_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3848_cast_fp16 = softmax(axis = var_3702, x = aw_501_cast_fp16)[name = tensor<string, []>("op_3848_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3849_cast_fp16 = softmax(axis = var_3702, x = aw_503_cast_fp16)[name = tensor<string, []>("op_3849_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3850_cast_fp16 = softmax(axis = var_3702, x = aw_505_cast_fp16)[name = tensor<string, []>("op_3850_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3851_cast_fp16 = softmax(axis = var_3702, x = aw_507_cast_fp16)[name = tensor<string, []>("op_3851_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3852_cast_fp16 = softmax(axis = var_3702, x = aw_509_cast_fp16)[name = tensor<string, []>("op_3852_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_3853_cast_fp16 = softmax(axis = var_3702, x = aw_511_cast_fp16)[name = tensor<string, []>("op_3853_cast_fp16")];
            tensor<string, []> var_3855_equation_0 = const()[name = tensor<string, []>("op_3855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3855_cast_fp16 = einsum(equation = var_3855_equation_0, values = (var_3789_cast_fp16_0, var_3838_cast_fp16))[name = tensor<string, []>("op_3855_cast_fp16")];
            tensor<string, []> var_3857_equation_0 = const()[name = tensor<string, []>("op_3857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3857_cast_fp16 = einsum(equation = var_3857_equation_0, values = (var_3789_cast_fp16_1, var_3839_cast_fp16))[name = tensor<string, []>("op_3857_cast_fp16")];
            tensor<string, []> var_3859_equation_0 = const()[name = tensor<string, []>("op_3859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3859_cast_fp16 = einsum(equation = var_3859_equation_0, values = (var_3789_cast_fp16_2, var_3840_cast_fp16))[name = tensor<string, []>("op_3859_cast_fp16")];
            tensor<string, []> var_3861_equation_0 = const()[name = tensor<string, []>("op_3861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3861_cast_fp16 = einsum(equation = var_3861_equation_0, values = (var_3789_cast_fp16_3, var_3841_cast_fp16))[name = tensor<string, []>("op_3861_cast_fp16")];
            tensor<string, []> var_3863_equation_0 = const()[name = tensor<string, []>("op_3863_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3863_cast_fp16 = einsum(equation = var_3863_equation_0, values = (var_3789_cast_fp16_4, var_3842_cast_fp16))[name = tensor<string, []>("op_3863_cast_fp16")];
            tensor<string, []> var_3865_equation_0 = const()[name = tensor<string, []>("op_3865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3865_cast_fp16 = einsum(equation = var_3865_equation_0, values = (var_3789_cast_fp16_5, var_3843_cast_fp16))[name = tensor<string, []>("op_3865_cast_fp16")];
            tensor<string, []> var_3867_equation_0 = const()[name = tensor<string, []>("op_3867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3867_cast_fp16 = einsum(equation = var_3867_equation_0, values = (var_3789_cast_fp16_6, var_3844_cast_fp16))[name = tensor<string, []>("op_3867_cast_fp16")];
            tensor<string, []> var_3869_equation_0 = const()[name = tensor<string, []>("op_3869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3869_cast_fp16 = einsum(equation = var_3869_equation_0, values = (var_3789_cast_fp16_7, var_3845_cast_fp16))[name = tensor<string, []>("op_3869_cast_fp16")];
            tensor<string, []> var_3871_equation_0 = const()[name = tensor<string, []>("op_3871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3871_cast_fp16 = einsum(equation = var_3871_equation_0, values = (var_3789_cast_fp16_8, var_3846_cast_fp16))[name = tensor<string, []>("op_3871_cast_fp16")];
            tensor<string, []> var_3873_equation_0 = const()[name = tensor<string, []>("op_3873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3873_cast_fp16 = einsum(equation = var_3873_equation_0, values = (var_3789_cast_fp16_9, var_3847_cast_fp16))[name = tensor<string, []>("op_3873_cast_fp16")];
            tensor<string, []> var_3875_equation_0 = const()[name = tensor<string, []>("op_3875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3875_cast_fp16 = einsum(equation = var_3875_equation_0, values = (var_3789_cast_fp16_10, var_3848_cast_fp16))[name = tensor<string, []>("op_3875_cast_fp16")];
            tensor<string, []> var_3877_equation_0 = const()[name = tensor<string, []>("op_3877_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3877_cast_fp16 = einsum(equation = var_3877_equation_0, values = (var_3789_cast_fp16_11, var_3849_cast_fp16))[name = tensor<string, []>("op_3877_cast_fp16")];
            tensor<string, []> var_3879_equation_0 = const()[name = tensor<string, []>("op_3879_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3879_cast_fp16 = einsum(equation = var_3879_equation_0, values = (var_3789_cast_fp16_12, var_3850_cast_fp16))[name = tensor<string, []>("op_3879_cast_fp16")];
            tensor<string, []> var_3881_equation_0 = const()[name = tensor<string, []>("op_3881_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3881_cast_fp16 = einsum(equation = var_3881_equation_0, values = (var_3789_cast_fp16_13, var_3851_cast_fp16))[name = tensor<string, []>("op_3881_cast_fp16")];
            tensor<string, []> var_3883_equation_0 = const()[name = tensor<string, []>("op_3883_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3883_cast_fp16 = einsum(equation = var_3883_equation_0, values = (var_3789_cast_fp16_14, var_3852_cast_fp16))[name = tensor<string, []>("op_3883_cast_fp16")];
            tensor<string, []> var_3885_equation_0 = const()[name = tensor<string, []>("op_3885_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_3885_cast_fp16 = einsum(equation = var_3885_equation_0, values = (var_3789_cast_fp16_15, var_3853_cast_fp16))[name = tensor<string, []>("op_3885_cast_fp16")];
            tensor<bool, []> input_157_interleave_0 = const()[name = tensor<string, []>("input_157_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_157_cast_fp16 = concat(axis = var_3702, interleave = input_157_interleave_0, values = (var_3855_cast_fp16, var_3857_cast_fp16, var_3859_cast_fp16, var_3861_cast_fp16, var_3863_cast_fp16, var_3865_cast_fp16, var_3867_cast_fp16, var_3869_cast_fp16, var_3871_cast_fp16, var_3873_cast_fp16, var_3875_cast_fp16, var_3877_cast_fp16, var_3879_cast_fp16, var_3881_cast_fp16, var_3883_cast_fp16, var_3885_cast_fp16))[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<string, []> var_3894_pad_type_0 = const()[name = tensor<string, []>("op_3894_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3894_strides_0 = const()[name = tensor<string, []>("op_3894_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3894_pad_0 = const()[name = tensor<string, []>("op_3894_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3894_dilations_0 = const()[name = tensor<string, []>("op_3894_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3894_groups_0 = const()[name = tensor<string, []>("op_3894_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_15_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98356096))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98880448))), name = tensor<string, []>("blocks_15_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_15_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98880576)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3894_cast_fp16 = conv(bias = blocks_15_attn_out_bias_to_fp16, dilations = var_3894_dilations_0, groups = var_3894_groups_0, pad = var_3894_pad_0, pad_type = var_3894_pad_type_0, strides = var_3894_strides_0, weight = blocks_15_attn_out_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor<string, []>("op_3894_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = var_3894_cast_fp16)[name = tensor<string, []>("inputs_63_cast_fp16")];
            tensor<int32, [1]> input_159_axes_0 = const()[name = tensor<string, []>("input_159_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_159_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_159_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98882688)))];
            tensor<fp16, [1024]> input_159_beta_0_to_fp16 = const()[name = tensor<string, []>("input_159_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98884800)))];
            tensor<fp16, []> var_3904_to_fp16 = const()[name = tensor<string, []>("op_3904_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_159_cast_fp16 = layer_norm(axes = input_159_axes_0, beta = input_159_beta_0_to_fp16, epsilon = var_3904_to_fp16, gamma = input_159_gamma_0_to_fp16, x = inputs_63_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")];
            tensor<string, []> input_161_pad_type_0 = const()[name = tensor<string, []>("input_161_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_161_strides_0 = const()[name = tensor<string, []>("input_161_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_161_pad_0 = const()[name = tensor<string, []>("input_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_161_dilations_0 = const()[name = tensor<string, []>("input_161_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_161_groups_0 = const()[name = tensor<string, []>("input_161_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_15_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98886912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100984128))), name = tensor<string, []>("blocks_15_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_15_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100984256)))];
            tensor<fp16, [1, 4096, 1, 600]> input_161_cast_fp16 = conv(bias = blocks_15_mlp_0_bias_to_fp16, dilations = input_161_dilations_0, groups = input_161_groups_0, pad = input_161_pad_0, pad_type = input_161_pad_type_0, strides = input_161_strides_0, weight = blocks_15_mlp_0_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<string, []> input_163_mode_0 = const()[name = tensor<string, []>("input_163_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_163_cast_fp16 = gelu(mode = input_163_mode_0, x = input_161_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<string, []> var_3930_pad_type_0 = const()[name = tensor<string, []>("op_3930_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3930_strides_0 = const()[name = tensor<string, []>("op_3930_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3930_pad_0 = const()[name = tensor<string, []>("op_3930_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3930_dilations_0 = const()[name = tensor<string, []>("op_3930_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3930_groups_0 = const()[name = tensor<string, []>("op_3930_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_15_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100992512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103089728))), name = tensor<string, []>("blocks_15_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_15_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103089856)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3930_cast_fp16 = conv(bias = blocks_15_mlp_2_bias_to_fp16, dilations = var_3930_dilations_0, groups = var_3930_groups_0, pad = var_3930_pad_0, pad_type = var_3930_pad_type_0, strides = var_3930_strides_0, weight = blocks_15_mlp_2_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = tensor<string, []>("op_3930_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = var_3930_cast_fp16)[name = tensor<string, []>("inputs_65_cast_fp16")];
            tensor<int32, []> var_3939 = const()[name = tensor<string, []>("op_3939"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_165_axes_0 = const()[name = tensor<string, []>("input_165_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_165_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_165_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103091968)))];
            tensor<fp16, [1024]> input_165_beta_0_to_fp16 = const()[name = tensor<string, []>("input_165_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103094080)))];
            tensor<fp16, []> var_3955_to_fp16 = const()[name = tensor<string, []>("op_3955_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = input_165_beta_0_to_fp16, epsilon = var_3955_to_fp16, gamma = input_165_gamma_0_to_fp16, x = inputs_65_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<string, []> q_33_pad_type_0 = const()[name = tensor<string, []>("q_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_33_strides_0 = const()[name = tensor<string, []>("q_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_33_pad_0 = const()[name = tensor<string, []>("q_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_33_dilations_0 = const()[name = tensor<string, []>("q_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_33_groups_0 = const()[name = tensor<string, []>("q_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_3990_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103096192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103620544))), name = tensor<string, []>("op_3990_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_3990_bias_0_to_fp16 = const()[name = tensor<string, []>("op_3990_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103620672)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3990_cast_fp16 = conv(bias = var_3990_bias_0_to_fp16, dilations = q_33_dilations_0, groups = q_33_groups_0, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = q_33_strides_0, weight = op_3990_weight_0_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor<string, []>("op_3990_cast_fp16")];
            tensor<string, []> k_33_pad_type_0 = const()[name = tensor<string, []>("k_33_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_33_strides_0 = const()[name = tensor<string, []>("k_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_33_pad_0 = const()[name = tensor<string, []>("k_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_33_dilations_0 = const()[name = tensor<string, []>("k_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_33_groups_0 = const()[name = tensor<string, []>("k_33_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_16_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103622784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104147136))), name = tensor<string, []>("blocks_16_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_33_cast_fp16 = conv(dilations = k_33_dilations_0, groups = k_33_groups_0, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = k_33_strides_0, weight = blocks_16_attn_key_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor<string, []>("k_33_cast_fp16")];
            tensor<string, []> var_3988_pad_type_0 = const()[name = tensor<string, []>("op_3988_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_3988_strides_0 = const()[name = tensor<string, []>("op_3988_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3988_pad_0 = const()[name = tensor<string, []>("op_3988_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3988_dilations_0 = const()[name = tensor<string, []>("op_3988_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_3988_groups_0 = const()[name = tensor<string, []>("op_3988_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_16_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104147264))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104671616))), name = tensor<string, []>("blocks_16_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_16_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104671744)))];
            tensor<fp16, [1, 1024, 1, 600]> var_3988_cast_fp16 = conv(bias = blocks_16_attn_value_bias_to_fp16, dilations = var_3988_dilations_0, groups = var_3988_groups_0, pad = var_3988_pad_0, pad_type = var_3988_pad_type_0, strides = var_3988_strides_0, weight = blocks_16_attn_value_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor<string, []>("op_3988_cast_fp16")];
            tensor<int32, [16]> tile_48 = const()[name = tensor<string, []>("tile_48"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_3991_axis_0 = const()[name = tensor<string, []>("op_3991_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_3991_cast_fp16_15 = split(axis = var_3991_axis_0, split_sizes = tile_48, x = var_3990_cast_fp16)[name = tensor<string, []>("op_3991_cast_fp16")];
            tensor<int32, [4]> var_4008_perm_0 = const()[name = tensor<string, []>("op_4008_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_49 = const()[name = tensor<string, []>("tile_49"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4009_axis_0 = const()[name = tensor<string, []>("op_4009_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_4008_cast_fp16 = transpose(perm = var_4008_perm_0, x = k_33_cast_fp16)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_4009_cast_fp16_15 = split(axis = var_4009_axis_0, split_sizes = tile_49, x = var_4008_cast_fp16)[name = tensor<string, []>("op_4009_cast_fp16")];
            tensor<int32, [16]> tile_50 = const()[name = tensor<string, []>("tile_50"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4026_axis_0 = const()[name = tensor<string, []>("op_4026_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4026_cast_fp16_15 = split(axis = var_4026_axis_0, split_sizes = tile_50, x = var_3988_cast_fp16)[name = tensor<string, []>("op_4026_cast_fp16")];
            tensor<string, []> aw_513_equation_0 = const()[name = tensor<string, []>("aw_513_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_513_cast_fp16 = einsum(equation = aw_513_equation_0, values = (var_4009_cast_fp16_0, var_3991_cast_fp16_0))[name = tensor<string, []>("aw_513_cast_fp16")];
            tensor<string, []> aw_515_equation_0 = const()[name = tensor<string, []>("aw_515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_515_cast_fp16 = einsum(equation = aw_515_equation_0, values = (var_4009_cast_fp16_1, var_3991_cast_fp16_1))[name = tensor<string, []>("aw_515_cast_fp16")];
            tensor<string, []> aw_517_equation_0 = const()[name = tensor<string, []>("aw_517_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_517_cast_fp16 = einsum(equation = aw_517_equation_0, values = (var_4009_cast_fp16_2, var_3991_cast_fp16_2))[name = tensor<string, []>("aw_517_cast_fp16")];
            tensor<string, []> aw_519_equation_0 = const()[name = tensor<string, []>("aw_519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_519_cast_fp16 = einsum(equation = aw_519_equation_0, values = (var_4009_cast_fp16_3, var_3991_cast_fp16_3))[name = tensor<string, []>("aw_519_cast_fp16")];
            tensor<string, []> aw_521_equation_0 = const()[name = tensor<string, []>("aw_521_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_521_cast_fp16 = einsum(equation = aw_521_equation_0, values = (var_4009_cast_fp16_4, var_3991_cast_fp16_4))[name = tensor<string, []>("aw_521_cast_fp16")];
            tensor<string, []> aw_523_equation_0 = const()[name = tensor<string, []>("aw_523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_523_cast_fp16 = einsum(equation = aw_523_equation_0, values = (var_4009_cast_fp16_5, var_3991_cast_fp16_5))[name = tensor<string, []>("aw_523_cast_fp16")];
            tensor<string, []> aw_525_equation_0 = const()[name = tensor<string, []>("aw_525_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_525_cast_fp16 = einsum(equation = aw_525_equation_0, values = (var_4009_cast_fp16_6, var_3991_cast_fp16_6))[name = tensor<string, []>("aw_525_cast_fp16")];
            tensor<string, []> aw_527_equation_0 = const()[name = tensor<string, []>("aw_527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_527_cast_fp16 = einsum(equation = aw_527_equation_0, values = (var_4009_cast_fp16_7, var_3991_cast_fp16_7))[name = tensor<string, []>("aw_527_cast_fp16")];
            tensor<string, []> aw_529_equation_0 = const()[name = tensor<string, []>("aw_529_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_529_cast_fp16 = einsum(equation = aw_529_equation_0, values = (var_4009_cast_fp16_8, var_3991_cast_fp16_8))[name = tensor<string, []>("aw_529_cast_fp16")];
            tensor<string, []> aw_531_equation_0 = const()[name = tensor<string, []>("aw_531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_531_cast_fp16 = einsum(equation = aw_531_equation_0, values = (var_4009_cast_fp16_9, var_3991_cast_fp16_9))[name = tensor<string, []>("aw_531_cast_fp16")];
            tensor<string, []> aw_533_equation_0 = const()[name = tensor<string, []>("aw_533_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_533_cast_fp16 = einsum(equation = aw_533_equation_0, values = (var_4009_cast_fp16_10, var_3991_cast_fp16_10))[name = tensor<string, []>("aw_533_cast_fp16")];
            tensor<string, []> aw_535_equation_0 = const()[name = tensor<string, []>("aw_535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_535_cast_fp16 = einsum(equation = aw_535_equation_0, values = (var_4009_cast_fp16_11, var_3991_cast_fp16_11))[name = tensor<string, []>("aw_535_cast_fp16")];
            tensor<string, []> aw_537_equation_0 = const()[name = tensor<string, []>("aw_537_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_537_cast_fp16 = einsum(equation = aw_537_equation_0, values = (var_4009_cast_fp16_12, var_3991_cast_fp16_12))[name = tensor<string, []>("aw_537_cast_fp16")];
            tensor<string, []> aw_539_equation_0 = const()[name = tensor<string, []>("aw_539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_539_cast_fp16 = einsum(equation = aw_539_equation_0, values = (var_4009_cast_fp16_13, var_3991_cast_fp16_13))[name = tensor<string, []>("aw_539_cast_fp16")];
            tensor<string, []> aw_541_equation_0 = const()[name = tensor<string, []>("aw_541_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_541_cast_fp16 = einsum(equation = aw_541_equation_0, values = (var_4009_cast_fp16_14, var_3991_cast_fp16_14))[name = tensor<string, []>("aw_541_cast_fp16")];
            tensor<string, []> aw_543_equation_0 = const()[name = tensor<string, []>("aw_543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_543_cast_fp16 = einsum(equation = aw_543_equation_0, values = (var_4009_cast_fp16_15, var_3991_cast_fp16_15))[name = tensor<string, []>("aw_543_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4075_cast_fp16 = softmax(axis = var_3939, x = aw_513_cast_fp16)[name = tensor<string, []>("op_4075_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4076_cast_fp16 = softmax(axis = var_3939, x = aw_515_cast_fp16)[name = tensor<string, []>("op_4076_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4077_cast_fp16 = softmax(axis = var_3939, x = aw_517_cast_fp16)[name = tensor<string, []>("op_4077_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4078_cast_fp16 = softmax(axis = var_3939, x = aw_519_cast_fp16)[name = tensor<string, []>("op_4078_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4079_cast_fp16 = softmax(axis = var_3939, x = aw_521_cast_fp16)[name = tensor<string, []>("op_4079_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4080_cast_fp16 = softmax(axis = var_3939, x = aw_523_cast_fp16)[name = tensor<string, []>("op_4080_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4081_cast_fp16 = softmax(axis = var_3939, x = aw_525_cast_fp16)[name = tensor<string, []>("op_4081_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4082_cast_fp16 = softmax(axis = var_3939, x = aw_527_cast_fp16)[name = tensor<string, []>("op_4082_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4083_cast_fp16 = softmax(axis = var_3939, x = aw_529_cast_fp16)[name = tensor<string, []>("op_4083_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4084_cast_fp16 = softmax(axis = var_3939, x = aw_531_cast_fp16)[name = tensor<string, []>("op_4084_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4085_cast_fp16 = softmax(axis = var_3939, x = aw_533_cast_fp16)[name = tensor<string, []>("op_4085_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4086_cast_fp16 = softmax(axis = var_3939, x = aw_535_cast_fp16)[name = tensor<string, []>("op_4086_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4087_cast_fp16 = softmax(axis = var_3939, x = aw_537_cast_fp16)[name = tensor<string, []>("op_4087_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4088_cast_fp16 = softmax(axis = var_3939, x = aw_539_cast_fp16)[name = tensor<string, []>("op_4088_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4089_cast_fp16 = softmax(axis = var_3939, x = aw_541_cast_fp16)[name = tensor<string, []>("op_4089_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4090_cast_fp16 = softmax(axis = var_3939, x = aw_543_cast_fp16)[name = tensor<string, []>("op_4090_cast_fp16")];
            tensor<string, []> var_4092_equation_0 = const()[name = tensor<string, []>("op_4092_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4092_cast_fp16 = einsum(equation = var_4092_equation_0, values = (var_4026_cast_fp16_0, var_4075_cast_fp16))[name = tensor<string, []>("op_4092_cast_fp16")];
            tensor<string, []> var_4094_equation_0 = const()[name = tensor<string, []>("op_4094_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4094_cast_fp16 = einsum(equation = var_4094_equation_0, values = (var_4026_cast_fp16_1, var_4076_cast_fp16))[name = tensor<string, []>("op_4094_cast_fp16")];
            tensor<string, []> var_4096_equation_0 = const()[name = tensor<string, []>("op_4096_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4096_cast_fp16 = einsum(equation = var_4096_equation_0, values = (var_4026_cast_fp16_2, var_4077_cast_fp16))[name = tensor<string, []>("op_4096_cast_fp16")];
            tensor<string, []> var_4098_equation_0 = const()[name = tensor<string, []>("op_4098_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4098_cast_fp16 = einsum(equation = var_4098_equation_0, values = (var_4026_cast_fp16_3, var_4078_cast_fp16))[name = tensor<string, []>("op_4098_cast_fp16")];
            tensor<string, []> var_4100_equation_0 = const()[name = tensor<string, []>("op_4100_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4100_cast_fp16 = einsum(equation = var_4100_equation_0, values = (var_4026_cast_fp16_4, var_4079_cast_fp16))[name = tensor<string, []>("op_4100_cast_fp16")];
            tensor<string, []> var_4102_equation_0 = const()[name = tensor<string, []>("op_4102_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4102_cast_fp16 = einsum(equation = var_4102_equation_0, values = (var_4026_cast_fp16_5, var_4080_cast_fp16))[name = tensor<string, []>("op_4102_cast_fp16")];
            tensor<string, []> var_4104_equation_0 = const()[name = tensor<string, []>("op_4104_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4104_cast_fp16 = einsum(equation = var_4104_equation_0, values = (var_4026_cast_fp16_6, var_4081_cast_fp16))[name = tensor<string, []>("op_4104_cast_fp16")];
            tensor<string, []> var_4106_equation_0 = const()[name = tensor<string, []>("op_4106_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4106_cast_fp16 = einsum(equation = var_4106_equation_0, values = (var_4026_cast_fp16_7, var_4082_cast_fp16))[name = tensor<string, []>("op_4106_cast_fp16")];
            tensor<string, []> var_4108_equation_0 = const()[name = tensor<string, []>("op_4108_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4108_cast_fp16 = einsum(equation = var_4108_equation_0, values = (var_4026_cast_fp16_8, var_4083_cast_fp16))[name = tensor<string, []>("op_4108_cast_fp16")];
            tensor<string, []> var_4110_equation_0 = const()[name = tensor<string, []>("op_4110_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4110_cast_fp16 = einsum(equation = var_4110_equation_0, values = (var_4026_cast_fp16_9, var_4084_cast_fp16))[name = tensor<string, []>("op_4110_cast_fp16")];
            tensor<string, []> var_4112_equation_0 = const()[name = tensor<string, []>("op_4112_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4112_cast_fp16 = einsum(equation = var_4112_equation_0, values = (var_4026_cast_fp16_10, var_4085_cast_fp16))[name = tensor<string, []>("op_4112_cast_fp16")];
            tensor<string, []> var_4114_equation_0 = const()[name = tensor<string, []>("op_4114_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4114_cast_fp16 = einsum(equation = var_4114_equation_0, values = (var_4026_cast_fp16_11, var_4086_cast_fp16))[name = tensor<string, []>("op_4114_cast_fp16")];
            tensor<string, []> var_4116_equation_0 = const()[name = tensor<string, []>("op_4116_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4116_cast_fp16 = einsum(equation = var_4116_equation_0, values = (var_4026_cast_fp16_12, var_4087_cast_fp16))[name = tensor<string, []>("op_4116_cast_fp16")];
            tensor<string, []> var_4118_equation_0 = const()[name = tensor<string, []>("op_4118_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4118_cast_fp16 = einsum(equation = var_4118_equation_0, values = (var_4026_cast_fp16_13, var_4088_cast_fp16))[name = tensor<string, []>("op_4118_cast_fp16")];
            tensor<string, []> var_4120_equation_0 = const()[name = tensor<string, []>("op_4120_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4120_cast_fp16 = einsum(equation = var_4120_equation_0, values = (var_4026_cast_fp16_14, var_4089_cast_fp16))[name = tensor<string, []>("op_4120_cast_fp16")];
            tensor<string, []> var_4122_equation_0 = const()[name = tensor<string, []>("op_4122_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4122_cast_fp16 = einsum(equation = var_4122_equation_0, values = (var_4026_cast_fp16_15, var_4090_cast_fp16))[name = tensor<string, []>("op_4122_cast_fp16")];
            tensor<bool, []> input_167_interleave_0 = const()[name = tensor<string, []>("input_167_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_167_cast_fp16 = concat(axis = var_3939, interleave = input_167_interleave_0, values = (var_4092_cast_fp16, var_4094_cast_fp16, var_4096_cast_fp16, var_4098_cast_fp16, var_4100_cast_fp16, var_4102_cast_fp16, var_4104_cast_fp16, var_4106_cast_fp16, var_4108_cast_fp16, var_4110_cast_fp16, var_4112_cast_fp16, var_4114_cast_fp16, var_4116_cast_fp16, var_4118_cast_fp16, var_4120_cast_fp16, var_4122_cast_fp16))[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<string, []> var_4131_pad_type_0 = const()[name = tensor<string, []>("op_4131_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4131_strides_0 = const()[name = tensor<string, []>("op_4131_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4131_pad_0 = const()[name = tensor<string, []>("op_4131_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4131_dilations_0 = const()[name = tensor<string, []>("op_4131_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4131_groups_0 = const()[name = tensor<string, []>("op_4131_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_16_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104673856))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105198208))), name = tensor<string, []>("blocks_16_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_16_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105198336)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4131_cast_fp16 = conv(bias = blocks_16_attn_out_bias_to_fp16, dilations = var_4131_dilations_0, groups = var_4131_groups_0, pad = var_4131_pad_0, pad_type = var_4131_pad_type_0, strides = var_4131_strides_0, weight = blocks_16_attn_out_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor<string, []>("op_4131_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = var_4131_cast_fp16)[name = tensor<string, []>("inputs_67_cast_fp16")];
            tensor<int32, [1]> input_169_axes_0 = const()[name = tensor<string, []>("input_169_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_169_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_169_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105200448)))];
            tensor<fp16, [1024]> input_169_beta_0_to_fp16 = const()[name = tensor<string, []>("input_169_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105202560)))];
            tensor<fp16, []> var_4141_to_fp16 = const()[name = tensor<string, []>("op_4141_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_169_cast_fp16 = layer_norm(axes = input_169_axes_0, beta = input_169_beta_0_to_fp16, epsilon = var_4141_to_fp16, gamma = input_169_gamma_0_to_fp16, x = inputs_67_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")];
            tensor<string, []> input_171_pad_type_0 = const()[name = tensor<string, []>("input_171_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_171_strides_0 = const()[name = tensor<string, []>("input_171_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_171_pad_0 = const()[name = tensor<string, []>("input_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_171_dilations_0 = const()[name = tensor<string, []>("input_171_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_171_groups_0 = const()[name = tensor<string, []>("input_171_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_16_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105204672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107301888))), name = tensor<string, []>("blocks_16_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_16_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107302016)))];
            tensor<fp16, [1, 4096, 1, 600]> input_171_cast_fp16 = conv(bias = blocks_16_mlp_0_bias_to_fp16, dilations = input_171_dilations_0, groups = input_171_groups_0, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = input_171_strides_0, weight = blocks_16_mlp_0_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")];
            tensor<string, []> input_173_mode_0 = const()[name = tensor<string, []>("input_173_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_173_cast_fp16 = gelu(mode = input_173_mode_0, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
            tensor<string, []> var_4167_pad_type_0 = const()[name = tensor<string, []>("op_4167_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4167_strides_0 = const()[name = tensor<string, []>("op_4167_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4167_pad_0 = const()[name = tensor<string, []>("op_4167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4167_dilations_0 = const()[name = tensor<string, []>("op_4167_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4167_groups_0 = const()[name = tensor<string, []>("op_4167_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_16_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107310272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109407488))), name = tensor<string, []>("blocks_16_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_16_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109407616)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4167_cast_fp16 = conv(bias = blocks_16_mlp_2_bias_to_fp16, dilations = var_4167_dilations_0, groups = var_4167_groups_0, pad = var_4167_pad_0, pad_type = var_4167_pad_type_0, strides = var_4167_strides_0, weight = blocks_16_mlp_2_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor<string, []>("op_4167_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = var_4167_cast_fp16)[name = tensor<string, []>("inputs_69_cast_fp16")];
            tensor<int32, []> var_4176 = const()[name = tensor<string, []>("op_4176"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_175_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_175_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109409728)))];
            tensor<fp16, [1024]> input_175_beta_0_to_fp16 = const()[name = tensor<string, []>("input_175_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109411840)))];
            tensor<fp16, []> var_4192_to_fp16 = const()[name = tensor<string, []>("op_4192_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = input_175_beta_0_to_fp16, epsilon = var_4192_to_fp16, gamma = input_175_gamma_0_to_fp16, x = inputs_69_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<string, []> q_35_pad_type_0 = const()[name = tensor<string, []>("q_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_35_strides_0 = const()[name = tensor<string, []>("q_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_35_pad_0 = const()[name = tensor<string, []>("q_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_35_dilations_0 = const()[name = tensor<string, []>("q_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_35_groups_0 = const()[name = tensor<string, []>("q_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_4227_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109413952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109938304))), name = tensor<string, []>("op_4227_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_4227_bias_0_to_fp16 = const()[name = tensor<string, []>("op_4227_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109938432)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4227_cast_fp16 = conv(bias = var_4227_bias_0_to_fp16, dilations = q_35_dilations_0, groups = q_35_groups_0, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = q_35_strides_0, weight = op_4227_weight_0_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor<string, []>("op_4227_cast_fp16")];
            tensor<string, []> k_35_pad_type_0 = const()[name = tensor<string, []>("k_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_35_strides_0 = const()[name = tensor<string, []>("k_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_35_pad_0 = const()[name = tensor<string, []>("k_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_35_dilations_0 = const()[name = tensor<string, []>("k_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_35_groups_0 = const()[name = tensor<string, []>("k_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_17_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109940544))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110464896))), name = tensor<string, []>("blocks_17_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_35_cast_fp16 = conv(dilations = k_35_dilations_0, groups = k_35_groups_0, pad = k_35_pad_0, pad_type = k_35_pad_type_0, strides = k_35_strides_0, weight = blocks_17_attn_key_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor<string, []>("k_35_cast_fp16")];
            tensor<string, []> var_4225_pad_type_0 = const()[name = tensor<string, []>("op_4225_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4225_strides_0 = const()[name = tensor<string, []>("op_4225_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4225_pad_0 = const()[name = tensor<string, []>("op_4225_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4225_dilations_0 = const()[name = tensor<string, []>("op_4225_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4225_groups_0 = const()[name = tensor<string, []>("op_4225_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_17_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110465024))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110989376))), name = tensor<string, []>("blocks_17_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_17_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110989504)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4225_cast_fp16 = conv(bias = blocks_17_attn_value_bias_to_fp16, dilations = var_4225_dilations_0, groups = var_4225_groups_0, pad = var_4225_pad_0, pad_type = var_4225_pad_type_0, strides = var_4225_strides_0, weight = blocks_17_attn_value_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = tensor<string, []>("op_4225_cast_fp16")];
            tensor<int32, [16]> tile_51 = const()[name = tensor<string, []>("tile_51"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4228_axis_0 = const()[name = tensor<string, []>("op_4228_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4228_cast_fp16_15 = split(axis = var_4228_axis_0, split_sizes = tile_51, x = var_4227_cast_fp16)[name = tensor<string, []>("op_4228_cast_fp16")];
            tensor<int32, [4]> var_4245_perm_0 = const()[name = tensor<string, []>("op_4245_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_52 = const()[name = tensor<string, []>("tile_52"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4246_axis_0 = const()[name = tensor<string, []>("op_4246_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_4245_cast_fp16 = transpose(perm = var_4245_perm_0, x = k_35_cast_fp16)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_4246_cast_fp16_15 = split(axis = var_4246_axis_0, split_sizes = tile_52, x = var_4245_cast_fp16)[name = tensor<string, []>("op_4246_cast_fp16")];
            tensor<int32, [16]> tile_53 = const()[name = tensor<string, []>("tile_53"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4263_axis_0 = const()[name = tensor<string, []>("op_4263_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4263_cast_fp16_15 = split(axis = var_4263_axis_0, split_sizes = tile_53, x = var_4225_cast_fp16)[name = tensor<string, []>("op_4263_cast_fp16")];
            tensor<string, []> aw_545_equation_0 = const()[name = tensor<string, []>("aw_545_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_545_cast_fp16 = einsum(equation = aw_545_equation_0, values = (var_4246_cast_fp16_0, var_4228_cast_fp16_0))[name = tensor<string, []>("aw_545_cast_fp16")];
            tensor<string, []> aw_547_equation_0 = const()[name = tensor<string, []>("aw_547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_547_cast_fp16 = einsum(equation = aw_547_equation_0, values = (var_4246_cast_fp16_1, var_4228_cast_fp16_1))[name = tensor<string, []>("aw_547_cast_fp16")];
            tensor<string, []> aw_549_equation_0 = const()[name = tensor<string, []>("aw_549_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_549_cast_fp16 = einsum(equation = aw_549_equation_0, values = (var_4246_cast_fp16_2, var_4228_cast_fp16_2))[name = tensor<string, []>("aw_549_cast_fp16")];
            tensor<string, []> aw_551_equation_0 = const()[name = tensor<string, []>("aw_551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_551_cast_fp16 = einsum(equation = aw_551_equation_0, values = (var_4246_cast_fp16_3, var_4228_cast_fp16_3))[name = tensor<string, []>("aw_551_cast_fp16")];
            tensor<string, []> aw_553_equation_0 = const()[name = tensor<string, []>("aw_553_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_553_cast_fp16 = einsum(equation = aw_553_equation_0, values = (var_4246_cast_fp16_4, var_4228_cast_fp16_4))[name = tensor<string, []>("aw_553_cast_fp16")];
            tensor<string, []> aw_555_equation_0 = const()[name = tensor<string, []>("aw_555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_555_cast_fp16 = einsum(equation = aw_555_equation_0, values = (var_4246_cast_fp16_5, var_4228_cast_fp16_5))[name = tensor<string, []>("aw_555_cast_fp16")];
            tensor<string, []> aw_557_equation_0 = const()[name = tensor<string, []>("aw_557_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_557_cast_fp16 = einsum(equation = aw_557_equation_0, values = (var_4246_cast_fp16_6, var_4228_cast_fp16_6))[name = tensor<string, []>("aw_557_cast_fp16")];
            tensor<string, []> aw_559_equation_0 = const()[name = tensor<string, []>("aw_559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_559_cast_fp16 = einsum(equation = aw_559_equation_0, values = (var_4246_cast_fp16_7, var_4228_cast_fp16_7))[name = tensor<string, []>("aw_559_cast_fp16")];
            tensor<string, []> aw_561_equation_0 = const()[name = tensor<string, []>("aw_561_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_561_cast_fp16 = einsum(equation = aw_561_equation_0, values = (var_4246_cast_fp16_8, var_4228_cast_fp16_8))[name = tensor<string, []>("aw_561_cast_fp16")];
            tensor<string, []> aw_563_equation_0 = const()[name = tensor<string, []>("aw_563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_563_cast_fp16 = einsum(equation = aw_563_equation_0, values = (var_4246_cast_fp16_9, var_4228_cast_fp16_9))[name = tensor<string, []>("aw_563_cast_fp16")];
            tensor<string, []> aw_565_equation_0 = const()[name = tensor<string, []>("aw_565_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_565_cast_fp16 = einsum(equation = aw_565_equation_0, values = (var_4246_cast_fp16_10, var_4228_cast_fp16_10))[name = tensor<string, []>("aw_565_cast_fp16")];
            tensor<string, []> aw_567_equation_0 = const()[name = tensor<string, []>("aw_567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_567_cast_fp16 = einsum(equation = aw_567_equation_0, values = (var_4246_cast_fp16_11, var_4228_cast_fp16_11))[name = tensor<string, []>("aw_567_cast_fp16")];
            tensor<string, []> aw_569_equation_0 = const()[name = tensor<string, []>("aw_569_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_569_cast_fp16 = einsum(equation = aw_569_equation_0, values = (var_4246_cast_fp16_12, var_4228_cast_fp16_12))[name = tensor<string, []>("aw_569_cast_fp16")];
            tensor<string, []> aw_571_equation_0 = const()[name = tensor<string, []>("aw_571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_571_cast_fp16 = einsum(equation = aw_571_equation_0, values = (var_4246_cast_fp16_13, var_4228_cast_fp16_13))[name = tensor<string, []>("aw_571_cast_fp16")];
            tensor<string, []> aw_573_equation_0 = const()[name = tensor<string, []>("aw_573_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_573_cast_fp16 = einsum(equation = aw_573_equation_0, values = (var_4246_cast_fp16_14, var_4228_cast_fp16_14))[name = tensor<string, []>("aw_573_cast_fp16")];
            tensor<string, []> aw_575_equation_0 = const()[name = tensor<string, []>("aw_575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_575_cast_fp16 = einsum(equation = aw_575_equation_0, values = (var_4246_cast_fp16_15, var_4228_cast_fp16_15))[name = tensor<string, []>("aw_575_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4312_cast_fp16 = softmax(axis = var_4176, x = aw_545_cast_fp16)[name = tensor<string, []>("op_4312_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4313_cast_fp16 = softmax(axis = var_4176, x = aw_547_cast_fp16)[name = tensor<string, []>("op_4313_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4314_cast_fp16 = softmax(axis = var_4176, x = aw_549_cast_fp16)[name = tensor<string, []>("op_4314_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4315_cast_fp16 = softmax(axis = var_4176, x = aw_551_cast_fp16)[name = tensor<string, []>("op_4315_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4316_cast_fp16 = softmax(axis = var_4176, x = aw_553_cast_fp16)[name = tensor<string, []>("op_4316_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4317_cast_fp16 = softmax(axis = var_4176, x = aw_555_cast_fp16)[name = tensor<string, []>("op_4317_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4318_cast_fp16 = softmax(axis = var_4176, x = aw_557_cast_fp16)[name = tensor<string, []>("op_4318_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4319_cast_fp16 = softmax(axis = var_4176, x = aw_559_cast_fp16)[name = tensor<string, []>("op_4319_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4320_cast_fp16 = softmax(axis = var_4176, x = aw_561_cast_fp16)[name = tensor<string, []>("op_4320_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4321_cast_fp16 = softmax(axis = var_4176, x = aw_563_cast_fp16)[name = tensor<string, []>("op_4321_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4322_cast_fp16 = softmax(axis = var_4176, x = aw_565_cast_fp16)[name = tensor<string, []>("op_4322_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4323_cast_fp16 = softmax(axis = var_4176, x = aw_567_cast_fp16)[name = tensor<string, []>("op_4323_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4324_cast_fp16 = softmax(axis = var_4176, x = aw_569_cast_fp16)[name = tensor<string, []>("op_4324_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4325_cast_fp16 = softmax(axis = var_4176, x = aw_571_cast_fp16)[name = tensor<string, []>("op_4325_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4326_cast_fp16 = softmax(axis = var_4176, x = aw_573_cast_fp16)[name = tensor<string, []>("op_4326_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4327_cast_fp16 = softmax(axis = var_4176, x = aw_575_cast_fp16)[name = tensor<string, []>("op_4327_cast_fp16")];
            tensor<string, []> var_4329_equation_0 = const()[name = tensor<string, []>("op_4329_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4329_cast_fp16 = einsum(equation = var_4329_equation_0, values = (var_4263_cast_fp16_0, var_4312_cast_fp16))[name = tensor<string, []>("op_4329_cast_fp16")];
            tensor<string, []> var_4331_equation_0 = const()[name = tensor<string, []>("op_4331_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4331_cast_fp16 = einsum(equation = var_4331_equation_0, values = (var_4263_cast_fp16_1, var_4313_cast_fp16))[name = tensor<string, []>("op_4331_cast_fp16")];
            tensor<string, []> var_4333_equation_0 = const()[name = tensor<string, []>("op_4333_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4333_cast_fp16 = einsum(equation = var_4333_equation_0, values = (var_4263_cast_fp16_2, var_4314_cast_fp16))[name = tensor<string, []>("op_4333_cast_fp16")];
            tensor<string, []> var_4335_equation_0 = const()[name = tensor<string, []>("op_4335_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4335_cast_fp16 = einsum(equation = var_4335_equation_0, values = (var_4263_cast_fp16_3, var_4315_cast_fp16))[name = tensor<string, []>("op_4335_cast_fp16")];
            tensor<string, []> var_4337_equation_0 = const()[name = tensor<string, []>("op_4337_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4337_cast_fp16 = einsum(equation = var_4337_equation_0, values = (var_4263_cast_fp16_4, var_4316_cast_fp16))[name = tensor<string, []>("op_4337_cast_fp16")];
            tensor<string, []> var_4339_equation_0 = const()[name = tensor<string, []>("op_4339_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4339_cast_fp16 = einsum(equation = var_4339_equation_0, values = (var_4263_cast_fp16_5, var_4317_cast_fp16))[name = tensor<string, []>("op_4339_cast_fp16")];
            tensor<string, []> var_4341_equation_0 = const()[name = tensor<string, []>("op_4341_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4341_cast_fp16 = einsum(equation = var_4341_equation_0, values = (var_4263_cast_fp16_6, var_4318_cast_fp16))[name = tensor<string, []>("op_4341_cast_fp16")];
            tensor<string, []> var_4343_equation_0 = const()[name = tensor<string, []>("op_4343_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4343_cast_fp16 = einsum(equation = var_4343_equation_0, values = (var_4263_cast_fp16_7, var_4319_cast_fp16))[name = tensor<string, []>("op_4343_cast_fp16")];
            tensor<string, []> var_4345_equation_0 = const()[name = tensor<string, []>("op_4345_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4345_cast_fp16 = einsum(equation = var_4345_equation_0, values = (var_4263_cast_fp16_8, var_4320_cast_fp16))[name = tensor<string, []>("op_4345_cast_fp16")];
            tensor<string, []> var_4347_equation_0 = const()[name = tensor<string, []>("op_4347_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4347_cast_fp16 = einsum(equation = var_4347_equation_0, values = (var_4263_cast_fp16_9, var_4321_cast_fp16))[name = tensor<string, []>("op_4347_cast_fp16")];
            tensor<string, []> var_4349_equation_0 = const()[name = tensor<string, []>("op_4349_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4349_cast_fp16 = einsum(equation = var_4349_equation_0, values = (var_4263_cast_fp16_10, var_4322_cast_fp16))[name = tensor<string, []>("op_4349_cast_fp16")];
            tensor<string, []> var_4351_equation_0 = const()[name = tensor<string, []>("op_4351_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4351_cast_fp16 = einsum(equation = var_4351_equation_0, values = (var_4263_cast_fp16_11, var_4323_cast_fp16))[name = tensor<string, []>("op_4351_cast_fp16")];
            tensor<string, []> var_4353_equation_0 = const()[name = tensor<string, []>("op_4353_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4353_cast_fp16 = einsum(equation = var_4353_equation_0, values = (var_4263_cast_fp16_12, var_4324_cast_fp16))[name = tensor<string, []>("op_4353_cast_fp16")];
            tensor<string, []> var_4355_equation_0 = const()[name = tensor<string, []>("op_4355_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4355_cast_fp16 = einsum(equation = var_4355_equation_0, values = (var_4263_cast_fp16_13, var_4325_cast_fp16))[name = tensor<string, []>("op_4355_cast_fp16")];
            tensor<string, []> var_4357_equation_0 = const()[name = tensor<string, []>("op_4357_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4357_cast_fp16 = einsum(equation = var_4357_equation_0, values = (var_4263_cast_fp16_14, var_4326_cast_fp16))[name = tensor<string, []>("op_4357_cast_fp16")];
            tensor<string, []> var_4359_equation_0 = const()[name = tensor<string, []>("op_4359_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4359_cast_fp16 = einsum(equation = var_4359_equation_0, values = (var_4263_cast_fp16_15, var_4327_cast_fp16))[name = tensor<string, []>("op_4359_cast_fp16")];
            tensor<bool, []> input_177_interleave_0 = const()[name = tensor<string, []>("input_177_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_177_cast_fp16 = concat(axis = var_4176, interleave = input_177_interleave_0, values = (var_4329_cast_fp16, var_4331_cast_fp16, var_4333_cast_fp16, var_4335_cast_fp16, var_4337_cast_fp16, var_4339_cast_fp16, var_4341_cast_fp16, var_4343_cast_fp16, var_4345_cast_fp16, var_4347_cast_fp16, var_4349_cast_fp16, var_4351_cast_fp16, var_4353_cast_fp16, var_4355_cast_fp16, var_4357_cast_fp16, var_4359_cast_fp16))[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<string, []> var_4368_pad_type_0 = const()[name = tensor<string, []>("op_4368_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4368_strides_0 = const()[name = tensor<string, []>("op_4368_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4368_pad_0 = const()[name = tensor<string, []>("op_4368_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4368_dilations_0 = const()[name = tensor<string, []>("op_4368_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4368_groups_0 = const()[name = tensor<string, []>("op_4368_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_17_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110991616))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111515968))), name = tensor<string, []>("blocks_17_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_17_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111516096)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4368_cast_fp16 = conv(bias = blocks_17_attn_out_bias_to_fp16, dilations = var_4368_dilations_0, groups = var_4368_groups_0, pad = var_4368_pad_0, pad_type = var_4368_pad_type_0, strides = var_4368_strides_0, weight = blocks_17_attn_out_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor<string, []>("op_4368_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = var_4368_cast_fp16)[name = tensor<string, []>("inputs_71_cast_fp16")];
            tensor<int32, [1]> input_179_axes_0 = const()[name = tensor<string, []>("input_179_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_179_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_179_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111518208)))];
            tensor<fp16, [1024]> input_179_beta_0_to_fp16 = const()[name = tensor<string, []>("input_179_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111520320)))];
            tensor<fp16, []> var_4378_to_fp16 = const()[name = tensor<string, []>("op_4378_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_179_cast_fp16 = layer_norm(axes = input_179_axes_0, beta = input_179_beta_0_to_fp16, epsilon = var_4378_to_fp16, gamma = input_179_gamma_0_to_fp16, x = inputs_71_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")];
            tensor<string, []> input_181_pad_type_0 = const()[name = tensor<string, []>("input_181_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_181_strides_0 = const()[name = tensor<string, []>("input_181_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_181_pad_0 = const()[name = tensor<string, []>("input_181_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_181_dilations_0 = const()[name = tensor<string, []>("input_181_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_181_groups_0 = const()[name = tensor<string, []>("input_181_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_17_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111522432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113619648))), name = tensor<string, []>("blocks_17_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_17_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113619776)))];
            tensor<fp16, [1, 4096, 1, 600]> input_181_cast_fp16 = conv(bias = blocks_17_mlp_0_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = blocks_17_mlp_0_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<string, []> input_183_mode_0 = const()[name = tensor<string, []>("input_183_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")];
            tensor<string, []> var_4404_pad_type_0 = const()[name = tensor<string, []>("op_4404_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4404_strides_0 = const()[name = tensor<string, []>("op_4404_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4404_pad_0 = const()[name = tensor<string, []>("op_4404_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4404_dilations_0 = const()[name = tensor<string, []>("op_4404_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4404_groups_0 = const()[name = tensor<string, []>("op_4404_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_17_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113628032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115725248))), name = tensor<string, []>("blocks_17_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_17_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115725376)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4404_cast_fp16 = conv(bias = blocks_17_mlp_2_bias_to_fp16, dilations = var_4404_dilations_0, groups = var_4404_groups_0, pad = var_4404_pad_0, pad_type = var_4404_pad_type_0, strides = var_4404_strides_0, weight = blocks_17_mlp_2_weight_to_fp16_palettized, x = input_183_cast_fp16)[name = tensor<string, []>("op_4404_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = var_4404_cast_fp16)[name = tensor<string, []>("inputs_73_cast_fp16")];
            tensor<int32, []> var_4413 = const()[name = tensor<string, []>("op_4413"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_185_axes_0 = const()[name = tensor<string, []>("input_185_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_185_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_185_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115727488)))];
            tensor<fp16, [1024]> input_185_beta_0_to_fp16 = const()[name = tensor<string, []>("input_185_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115729600)))];
            tensor<fp16, []> var_4429_to_fp16 = const()[name = tensor<string, []>("op_4429_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = input_185_beta_0_to_fp16, epsilon = var_4429_to_fp16, gamma = input_185_gamma_0_to_fp16, x = inputs_73_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<string, []> q_37_pad_type_0 = const()[name = tensor<string, []>("q_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_37_strides_0 = const()[name = tensor<string, []>("q_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_37_pad_0 = const()[name = tensor<string, []>("q_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_37_dilations_0 = const()[name = tensor<string, []>("q_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_37_groups_0 = const()[name = tensor<string, []>("q_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_4464_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115731712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116256064))), name = tensor<string, []>("op_4464_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_4464_bias_0_to_fp16 = const()[name = tensor<string, []>("op_4464_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116256192)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4464_cast_fp16 = conv(bias = var_4464_bias_0_to_fp16, dilations = q_37_dilations_0, groups = q_37_groups_0, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = q_37_strides_0, weight = op_4464_weight_0_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("op_4464_cast_fp16")];
            tensor<string, []> k_37_pad_type_0 = const()[name = tensor<string, []>("k_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_37_strides_0 = const()[name = tensor<string, []>("k_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_37_pad_0 = const()[name = tensor<string, []>("k_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_37_dilations_0 = const()[name = tensor<string, []>("k_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_37_groups_0 = const()[name = tensor<string, []>("k_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_18_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116258304))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116782656))), name = tensor<string, []>("blocks_18_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_37_cast_fp16 = conv(dilations = k_37_dilations_0, groups = k_37_groups_0, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = k_37_strides_0, weight = blocks_18_attn_key_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("k_37_cast_fp16")];
            tensor<string, []> var_4462_pad_type_0 = const()[name = tensor<string, []>("op_4462_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4462_strides_0 = const()[name = tensor<string, []>("op_4462_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4462_pad_0 = const()[name = tensor<string, []>("op_4462_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4462_dilations_0 = const()[name = tensor<string, []>("op_4462_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4462_groups_0 = const()[name = tensor<string, []>("op_4462_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_18_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116782784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117307136))), name = tensor<string, []>("blocks_18_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_18_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117307264)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4462_cast_fp16 = conv(bias = blocks_18_attn_value_bias_to_fp16, dilations = var_4462_dilations_0, groups = var_4462_groups_0, pad = var_4462_pad_0, pad_type = var_4462_pad_type_0, strides = var_4462_strides_0, weight = blocks_18_attn_value_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("op_4462_cast_fp16")];
            tensor<int32, [16]> tile_54 = const()[name = tensor<string, []>("tile_54"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4465_axis_0 = const()[name = tensor<string, []>("op_4465_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4465_cast_fp16_15 = split(axis = var_4465_axis_0, split_sizes = tile_54, x = var_4464_cast_fp16)[name = tensor<string, []>("op_4465_cast_fp16")];
            tensor<int32, [4]> var_4482_perm_0 = const()[name = tensor<string, []>("op_4482_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_55 = const()[name = tensor<string, []>("tile_55"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4483_axis_0 = const()[name = tensor<string, []>("op_4483_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_4482_cast_fp16 = transpose(perm = var_4482_perm_0, x = k_37_cast_fp16)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_4483_cast_fp16_15 = split(axis = var_4483_axis_0, split_sizes = tile_55, x = var_4482_cast_fp16)[name = tensor<string, []>("op_4483_cast_fp16")];
            tensor<int32, [16]> tile_56 = const()[name = tensor<string, []>("tile_56"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4500_axis_0 = const()[name = tensor<string, []>("op_4500_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4500_cast_fp16_15 = split(axis = var_4500_axis_0, split_sizes = tile_56, x = var_4462_cast_fp16)[name = tensor<string, []>("op_4500_cast_fp16")];
            tensor<string, []> aw_577_equation_0 = const()[name = tensor<string, []>("aw_577_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_577_cast_fp16 = einsum(equation = aw_577_equation_0, values = (var_4483_cast_fp16_0, var_4465_cast_fp16_0))[name = tensor<string, []>("aw_577_cast_fp16")];
            tensor<string, []> aw_579_equation_0 = const()[name = tensor<string, []>("aw_579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_579_cast_fp16 = einsum(equation = aw_579_equation_0, values = (var_4483_cast_fp16_1, var_4465_cast_fp16_1))[name = tensor<string, []>("aw_579_cast_fp16")];
            tensor<string, []> aw_581_equation_0 = const()[name = tensor<string, []>("aw_581_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_581_cast_fp16 = einsum(equation = aw_581_equation_0, values = (var_4483_cast_fp16_2, var_4465_cast_fp16_2))[name = tensor<string, []>("aw_581_cast_fp16")];
            tensor<string, []> aw_583_equation_0 = const()[name = tensor<string, []>("aw_583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_583_cast_fp16 = einsum(equation = aw_583_equation_0, values = (var_4483_cast_fp16_3, var_4465_cast_fp16_3))[name = tensor<string, []>("aw_583_cast_fp16")];
            tensor<string, []> aw_585_equation_0 = const()[name = tensor<string, []>("aw_585_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_585_cast_fp16 = einsum(equation = aw_585_equation_0, values = (var_4483_cast_fp16_4, var_4465_cast_fp16_4))[name = tensor<string, []>("aw_585_cast_fp16")];
            tensor<string, []> aw_587_equation_0 = const()[name = tensor<string, []>("aw_587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_587_cast_fp16 = einsum(equation = aw_587_equation_0, values = (var_4483_cast_fp16_5, var_4465_cast_fp16_5))[name = tensor<string, []>("aw_587_cast_fp16")];
            tensor<string, []> aw_589_equation_0 = const()[name = tensor<string, []>("aw_589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_589_cast_fp16 = einsum(equation = aw_589_equation_0, values = (var_4483_cast_fp16_6, var_4465_cast_fp16_6))[name = tensor<string, []>("aw_589_cast_fp16")];
            tensor<string, []> aw_591_equation_0 = const()[name = tensor<string, []>("aw_591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_591_cast_fp16 = einsum(equation = aw_591_equation_0, values = (var_4483_cast_fp16_7, var_4465_cast_fp16_7))[name = tensor<string, []>("aw_591_cast_fp16")];
            tensor<string, []> aw_593_equation_0 = const()[name = tensor<string, []>("aw_593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_593_cast_fp16 = einsum(equation = aw_593_equation_0, values = (var_4483_cast_fp16_8, var_4465_cast_fp16_8))[name = tensor<string, []>("aw_593_cast_fp16")];
            tensor<string, []> aw_595_equation_0 = const()[name = tensor<string, []>("aw_595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_595_cast_fp16 = einsum(equation = aw_595_equation_0, values = (var_4483_cast_fp16_9, var_4465_cast_fp16_9))[name = tensor<string, []>("aw_595_cast_fp16")];
            tensor<string, []> aw_597_equation_0 = const()[name = tensor<string, []>("aw_597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_597_cast_fp16 = einsum(equation = aw_597_equation_0, values = (var_4483_cast_fp16_10, var_4465_cast_fp16_10))[name = tensor<string, []>("aw_597_cast_fp16")];
            tensor<string, []> aw_599_equation_0 = const()[name = tensor<string, []>("aw_599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_599_cast_fp16 = einsum(equation = aw_599_equation_0, values = (var_4483_cast_fp16_11, var_4465_cast_fp16_11))[name = tensor<string, []>("aw_599_cast_fp16")];
            tensor<string, []> aw_601_equation_0 = const()[name = tensor<string, []>("aw_601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_601_cast_fp16 = einsum(equation = aw_601_equation_0, values = (var_4483_cast_fp16_12, var_4465_cast_fp16_12))[name = tensor<string, []>("aw_601_cast_fp16")];
            tensor<string, []> aw_603_equation_0 = const()[name = tensor<string, []>("aw_603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_603_cast_fp16 = einsum(equation = aw_603_equation_0, values = (var_4483_cast_fp16_13, var_4465_cast_fp16_13))[name = tensor<string, []>("aw_603_cast_fp16")];
            tensor<string, []> aw_605_equation_0 = const()[name = tensor<string, []>("aw_605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_605_cast_fp16 = einsum(equation = aw_605_equation_0, values = (var_4483_cast_fp16_14, var_4465_cast_fp16_14))[name = tensor<string, []>("aw_605_cast_fp16")];
            tensor<string, []> aw_607_equation_0 = const()[name = tensor<string, []>("aw_607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_607_cast_fp16 = einsum(equation = aw_607_equation_0, values = (var_4483_cast_fp16_15, var_4465_cast_fp16_15))[name = tensor<string, []>("aw_607_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4549_cast_fp16 = softmax(axis = var_4413, x = aw_577_cast_fp16)[name = tensor<string, []>("op_4549_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4550_cast_fp16 = softmax(axis = var_4413, x = aw_579_cast_fp16)[name = tensor<string, []>("op_4550_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4551_cast_fp16 = softmax(axis = var_4413, x = aw_581_cast_fp16)[name = tensor<string, []>("op_4551_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4552_cast_fp16 = softmax(axis = var_4413, x = aw_583_cast_fp16)[name = tensor<string, []>("op_4552_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4553_cast_fp16 = softmax(axis = var_4413, x = aw_585_cast_fp16)[name = tensor<string, []>("op_4553_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4554_cast_fp16 = softmax(axis = var_4413, x = aw_587_cast_fp16)[name = tensor<string, []>("op_4554_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4555_cast_fp16 = softmax(axis = var_4413, x = aw_589_cast_fp16)[name = tensor<string, []>("op_4555_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4556_cast_fp16 = softmax(axis = var_4413, x = aw_591_cast_fp16)[name = tensor<string, []>("op_4556_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4557_cast_fp16 = softmax(axis = var_4413, x = aw_593_cast_fp16)[name = tensor<string, []>("op_4557_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4558_cast_fp16 = softmax(axis = var_4413, x = aw_595_cast_fp16)[name = tensor<string, []>("op_4558_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4559_cast_fp16 = softmax(axis = var_4413, x = aw_597_cast_fp16)[name = tensor<string, []>("op_4559_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4560_cast_fp16 = softmax(axis = var_4413, x = aw_599_cast_fp16)[name = tensor<string, []>("op_4560_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4561_cast_fp16 = softmax(axis = var_4413, x = aw_601_cast_fp16)[name = tensor<string, []>("op_4561_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4562_cast_fp16 = softmax(axis = var_4413, x = aw_603_cast_fp16)[name = tensor<string, []>("op_4562_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4563_cast_fp16 = softmax(axis = var_4413, x = aw_605_cast_fp16)[name = tensor<string, []>("op_4563_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4564_cast_fp16 = softmax(axis = var_4413, x = aw_607_cast_fp16)[name = tensor<string, []>("op_4564_cast_fp16")];
            tensor<string, []> var_4566_equation_0 = const()[name = tensor<string, []>("op_4566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4566_cast_fp16 = einsum(equation = var_4566_equation_0, values = (var_4500_cast_fp16_0, var_4549_cast_fp16))[name = tensor<string, []>("op_4566_cast_fp16")];
            tensor<string, []> var_4568_equation_0 = const()[name = tensor<string, []>("op_4568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4568_cast_fp16 = einsum(equation = var_4568_equation_0, values = (var_4500_cast_fp16_1, var_4550_cast_fp16))[name = tensor<string, []>("op_4568_cast_fp16")];
            tensor<string, []> var_4570_equation_0 = const()[name = tensor<string, []>("op_4570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4570_cast_fp16 = einsum(equation = var_4570_equation_0, values = (var_4500_cast_fp16_2, var_4551_cast_fp16))[name = tensor<string, []>("op_4570_cast_fp16")];
            tensor<string, []> var_4572_equation_0 = const()[name = tensor<string, []>("op_4572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4572_cast_fp16 = einsum(equation = var_4572_equation_0, values = (var_4500_cast_fp16_3, var_4552_cast_fp16))[name = tensor<string, []>("op_4572_cast_fp16")];
            tensor<string, []> var_4574_equation_0 = const()[name = tensor<string, []>("op_4574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4574_cast_fp16 = einsum(equation = var_4574_equation_0, values = (var_4500_cast_fp16_4, var_4553_cast_fp16))[name = tensor<string, []>("op_4574_cast_fp16")];
            tensor<string, []> var_4576_equation_0 = const()[name = tensor<string, []>("op_4576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4576_cast_fp16 = einsum(equation = var_4576_equation_0, values = (var_4500_cast_fp16_5, var_4554_cast_fp16))[name = tensor<string, []>("op_4576_cast_fp16")];
            tensor<string, []> var_4578_equation_0 = const()[name = tensor<string, []>("op_4578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4578_cast_fp16 = einsum(equation = var_4578_equation_0, values = (var_4500_cast_fp16_6, var_4555_cast_fp16))[name = tensor<string, []>("op_4578_cast_fp16")];
            tensor<string, []> var_4580_equation_0 = const()[name = tensor<string, []>("op_4580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4580_cast_fp16 = einsum(equation = var_4580_equation_0, values = (var_4500_cast_fp16_7, var_4556_cast_fp16))[name = tensor<string, []>("op_4580_cast_fp16")];
            tensor<string, []> var_4582_equation_0 = const()[name = tensor<string, []>("op_4582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4582_cast_fp16 = einsum(equation = var_4582_equation_0, values = (var_4500_cast_fp16_8, var_4557_cast_fp16))[name = tensor<string, []>("op_4582_cast_fp16")];
            tensor<string, []> var_4584_equation_0 = const()[name = tensor<string, []>("op_4584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4584_cast_fp16 = einsum(equation = var_4584_equation_0, values = (var_4500_cast_fp16_9, var_4558_cast_fp16))[name = tensor<string, []>("op_4584_cast_fp16")];
            tensor<string, []> var_4586_equation_0 = const()[name = tensor<string, []>("op_4586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4586_cast_fp16 = einsum(equation = var_4586_equation_0, values = (var_4500_cast_fp16_10, var_4559_cast_fp16))[name = tensor<string, []>("op_4586_cast_fp16")];
            tensor<string, []> var_4588_equation_0 = const()[name = tensor<string, []>("op_4588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4588_cast_fp16 = einsum(equation = var_4588_equation_0, values = (var_4500_cast_fp16_11, var_4560_cast_fp16))[name = tensor<string, []>("op_4588_cast_fp16")];
            tensor<string, []> var_4590_equation_0 = const()[name = tensor<string, []>("op_4590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4590_cast_fp16 = einsum(equation = var_4590_equation_0, values = (var_4500_cast_fp16_12, var_4561_cast_fp16))[name = tensor<string, []>("op_4590_cast_fp16")];
            tensor<string, []> var_4592_equation_0 = const()[name = tensor<string, []>("op_4592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4592_cast_fp16 = einsum(equation = var_4592_equation_0, values = (var_4500_cast_fp16_13, var_4562_cast_fp16))[name = tensor<string, []>("op_4592_cast_fp16")];
            tensor<string, []> var_4594_equation_0 = const()[name = tensor<string, []>("op_4594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4594_cast_fp16 = einsum(equation = var_4594_equation_0, values = (var_4500_cast_fp16_14, var_4563_cast_fp16))[name = tensor<string, []>("op_4594_cast_fp16")];
            tensor<string, []> var_4596_equation_0 = const()[name = tensor<string, []>("op_4596_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4596_cast_fp16 = einsum(equation = var_4596_equation_0, values = (var_4500_cast_fp16_15, var_4564_cast_fp16))[name = tensor<string, []>("op_4596_cast_fp16")];
            tensor<bool, []> input_187_interleave_0 = const()[name = tensor<string, []>("input_187_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_187_cast_fp16 = concat(axis = var_4413, interleave = input_187_interleave_0, values = (var_4566_cast_fp16, var_4568_cast_fp16, var_4570_cast_fp16, var_4572_cast_fp16, var_4574_cast_fp16, var_4576_cast_fp16, var_4578_cast_fp16, var_4580_cast_fp16, var_4582_cast_fp16, var_4584_cast_fp16, var_4586_cast_fp16, var_4588_cast_fp16, var_4590_cast_fp16, var_4592_cast_fp16, var_4594_cast_fp16, var_4596_cast_fp16))[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<string, []> var_4605_pad_type_0 = const()[name = tensor<string, []>("op_4605_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4605_strides_0 = const()[name = tensor<string, []>("op_4605_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4605_pad_0 = const()[name = tensor<string, []>("op_4605_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4605_dilations_0 = const()[name = tensor<string, []>("op_4605_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4605_groups_0 = const()[name = tensor<string, []>("op_4605_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_18_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117309376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117833728))), name = tensor<string, []>("blocks_18_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_18_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117833856)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4605_cast_fp16 = conv(bias = blocks_18_attn_out_bias_to_fp16, dilations = var_4605_dilations_0, groups = var_4605_groups_0, pad = var_4605_pad_0, pad_type = var_4605_pad_type_0, strides = var_4605_strides_0, weight = blocks_18_attn_out_weight_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor<string, []>("op_4605_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = var_4605_cast_fp16)[name = tensor<string, []>("inputs_75_cast_fp16")];
            tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_189_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_189_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117835968)))];
            tensor<fp16, [1024]> input_189_beta_0_to_fp16 = const()[name = tensor<string, []>("input_189_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117838080)))];
            tensor<fp16, []> var_4615_to_fp16 = const()[name = tensor<string, []>("op_4615_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = input_189_beta_0_to_fp16, epsilon = var_4615_to_fp16, gamma = input_189_gamma_0_to_fp16, x = inputs_75_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<string, []> input_191_pad_type_0 = const()[name = tensor<string, []>("input_191_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_191_strides_0 = const()[name = tensor<string, []>("input_191_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_191_pad_0 = const()[name = tensor<string, []>("input_191_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_191_dilations_0 = const()[name = tensor<string, []>("input_191_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_191_groups_0 = const()[name = tensor<string, []>("input_191_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_18_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117840192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119937408))), name = tensor<string, []>("blocks_18_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_18_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119937536)))];
            tensor<fp16, [1, 4096, 1, 600]> input_191_cast_fp16 = conv(bias = blocks_18_mlp_0_bias_to_fp16, dilations = input_191_dilations_0, groups = input_191_groups_0, pad = input_191_pad_0, pad_type = input_191_pad_type_0, strides = input_191_strides_0, weight = blocks_18_mlp_0_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor<string, []>("input_191_cast_fp16")];
            tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = input_191_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<string, []> var_4641_pad_type_0 = const()[name = tensor<string, []>("op_4641_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4641_strides_0 = const()[name = tensor<string, []>("op_4641_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4641_pad_0 = const()[name = tensor<string, []>("op_4641_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4641_dilations_0 = const()[name = tensor<string, []>("op_4641_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4641_groups_0 = const()[name = tensor<string, []>("op_4641_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_18_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119945792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122043008))), name = tensor<string, []>("blocks_18_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_18_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122043136)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4641_cast_fp16 = conv(bias = blocks_18_mlp_2_bias_to_fp16, dilations = var_4641_dilations_0, groups = var_4641_groups_0, pad = var_4641_pad_0, pad_type = var_4641_pad_type_0, strides = var_4641_strides_0, weight = blocks_18_mlp_2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor<string, []>("op_4641_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = var_4641_cast_fp16)[name = tensor<string, []>("inputs_77_cast_fp16")];
            tensor<int32, []> var_4650 = const()[name = tensor<string, []>("op_4650"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_195_axes_0 = const()[name = tensor<string, []>("input_195_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_195_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_195_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122045248)))];
            tensor<fp16, [1024]> input_195_beta_0_to_fp16 = const()[name = tensor<string, []>("input_195_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122047360)))];
            tensor<fp16, []> var_4666_to_fp16 = const()[name = tensor<string, []>("op_4666_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_195_cast_fp16 = layer_norm(axes = input_195_axes_0, beta = input_195_beta_0_to_fp16, epsilon = var_4666_to_fp16, gamma = input_195_gamma_0_to_fp16, x = inputs_77_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<string, []> q_39_pad_type_0 = const()[name = tensor<string, []>("q_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_39_strides_0 = const()[name = tensor<string, []>("q_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_39_pad_0 = const()[name = tensor<string, []>("q_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_39_dilations_0 = const()[name = tensor<string, []>("q_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_39_groups_0 = const()[name = tensor<string, []>("q_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_4701_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122049472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122573824))), name = tensor<string, []>("op_4701_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_4701_bias_0_to_fp16 = const()[name = tensor<string, []>("op_4701_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122573952)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4701_cast_fp16 = conv(bias = var_4701_bias_0_to_fp16, dilations = q_39_dilations_0, groups = q_39_groups_0, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = q_39_strides_0, weight = op_4701_weight_0_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor<string, []>("op_4701_cast_fp16")];
            tensor<string, []> k_39_pad_type_0 = const()[name = tensor<string, []>("k_39_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_39_strides_0 = const()[name = tensor<string, []>("k_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_39_pad_0 = const()[name = tensor<string, []>("k_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_39_dilations_0 = const()[name = tensor<string, []>("k_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_39_groups_0 = const()[name = tensor<string, []>("k_39_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_19_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122576064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123100416))), name = tensor<string, []>("blocks_19_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_39_cast_fp16 = conv(dilations = k_39_dilations_0, groups = k_39_groups_0, pad = k_39_pad_0, pad_type = k_39_pad_type_0, strides = k_39_strides_0, weight = blocks_19_attn_key_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor<string, []>("k_39_cast_fp16")];
            tensor<string, []> var_4699_pad_type_0 = const()[name = tensor<string, []>("op_4699_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4699_strides_0 = const()[name = tensor<string, []>("op_4699_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4699_pad_0 = const()[name = tensor<string, []>("op_4699_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4699_dilations_0 = const()[name = tensor<string, []>("op_4699_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4699_groups_0 = const()[name = tensor<string, []>("op_4699_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_19_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123100544))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123624896))), name = tensor<string, []>("blocks_19_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_19_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123625024)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4699_cast_fp16 = conv(bias = blocks_19_attn_value_bias_to_fp16, dilations = var_4699_dilations_0, groups = var_4699_groups_0, pad = var_4699_pad_0, pad_type = var_4699_pad_type_0, strides = var_4699_strides_0, weight = blocks_19_attn_value_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor<string, []>("op_4699_cast_fp16")];
            tensor<int32, [16]> tile_57 = const()[name = tensor<string, []>("tile_57"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4702_axis_0 = const()[name = tensor<string, []>("op_4702_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4702_cast_fp16_15 = split(axis = var_4702_axis_0, split_sizes = tile_57, x = var_4701_cast_fp16)[name = tensor<string, []>("op_4702_cast_fp16")];
            tensor<int32, [4]> var_4719_perm_0 = const()[name = tensor<string, []>("op_4719_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_58 = const()[name = tensor<string, []>("tile_58"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4720_axis_0 = const()[name = tensor<string, []>("op_4720_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_4719_cast_fp16 = transpose(perm = var_4719_perm_0, x = k_39_cast_fp16)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_4720_cast_fp16_15 = split(axis = var_4720_axis_0, split_sizes = tile_58, x = var_4719_cast_fp16)[name = tensor<string, []>("op_4720_cast_fp16")];
            tensor<int32, [16]> tile_59 = const()[name = tensor<string, []>("tile_59"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4737_axis_0 = const()[name = tensor<string, []>("op_4737_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4737_cast_fp16_15 = split(axis = var_4737_axis_0, split_sizes = tile_59, x = var_4699_cast_fp16)[name = tensor<string, []>("op_4737_cast_fp16")];
            tensor<string, []> aw_609_equation_0 = const()[name = tensor<string, []>("aw_609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_609_cast_fp16 = einsum(equation = aw_609_equation_0, values = (var_4720_cast_fp16_0, var_4702_cast_fp16_0))[name = tensor<string, []>("aw_609_cast_fp16")];
            tensor<string, []> aw_611_equation_0 = const()[name = tensor<string, []>("aw_611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_611_cast_fp16 = einsum(equation = aw_611_equation_0, values = (var_4720_cast_fp16_1, var_4702_cast_fp16_1))[name = tensor<string, []>("aw_611_cast_fp16")];
            tensor<string, []> aw_613_equation_0 = const()[name = tensor<string, []>("aw_613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_613_cast_fp16 = einsum(equation = aw_613_equation_0, values = (var_4720_cast_fp16_2, var_4702_cast_fp16_2))[name = tensor<string, []>("aw_613_cast_fp16")];
            tensor<string, []> aw_615_equation_0 = const()[name = tensor<string, []>("aw_615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_615_cast_fp16 = einsum(equation = aw_615_equation_0, values = (var_4720_cast_fp16_3, var_4702_cast_fp16_3))[name = tensor<string, []>("aw_615_cast_fp16")];
            tensor<string, []> aw_617_equation_0 = const()[name = tensor<string, []>("aw_617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_617_cast_fp16 = einsum(equation = aw_617_equation_0, values = (var_4720_cast_fp16_4, var_4702_cast_fp16_4))[name = tensor<string, []>("aw_617_cast_fp16")];
            tensor<string, []> aw_619_equation_0 = const()[name = tensor<string, []>("aw_619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_619_cast_fp16 = einsum(equation = aw_619_equation_0, values = (var_4720_cast_fp16_5, var_4702_cast_fp16_5))[name = tensor<string, []>("aw_619_cast_fp16")];
            tensor<string, []> aw_621_equation_0 = const()[name = tensor<string, []>("aw_621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_621_cast_fp16 = einsum(equation = aw_621_equation_0, values = (var_4720_cast_fp16_6, var_4702_cast_fp16_6))[name = tensor<string, []>("aw_621_cast_fp16")];
            tensor<string, []> aw_623_equation_0 = const()[name = tensor<string, []>("aw_623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_623_cast_fp16 = einsum(equation = aw_623_equation_0, values = (var_4720_cast_fp16_7, var_4702_cast_fp16_7))[name = tensor<string, []>("aw_623_cast_fp16")];
            tensor<string, []> aw_625_equation_0 = const()[name = tensor<string, []>("aw_625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_625_cast_fp16 = einsum(equation = aw_625_equation_0, values = (var_4720_cast_fp16_8, var_4702_cast_fp16_8))[name = tensor<string, []>("aw_625_cast_fp16")];
            tensor<string, []> aw_627_equation_0 = const()[name = tensor<string, []>("aw_627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_627_cast_fp16 = einsum(equation = aw_627_equation_0, values = (var_4720_cast_fp16_9, var_4702_cast_fp16_9))[name = tensor<string, []>("aw_627_cast_fp16")];
            tensor<string, []> aw_629_equation_0 = const()[name = tensor<string, []>("aw_629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_629_cast_fp16 = einsum(equation = aw_629_equation_0, values = (var_4720_cast_fp16_10, var_4702_cast_fp16_10))[name = tensor<string, []>("aw_629_cast_fp16")];
            tensor<string, []> aw_631_equation_0 = const()[name = tensor<string, []>("aw_631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_631_cast_fp16 = einsum(equation = aw_631_equation_0, values = (var_4720_cast_fp16_11, var_4702_cast_fp16_11))[name = tensor<string, []>("aw_631_cast_fp16")];
            tensor<string, []> aw_633_equation_0 = const()[name = tensor<string, []>("aw_633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_633_cast_fp16 = einsum(equation = aw_633_equation_0, values = (var_4720_cast_fp16_12, var_4702_cast_fp16_12))[name = tensor<string, []>("aw_633_cast_fp16")];
            tensor<string, []> aw_635_equation_0 = const()[name = tensor<string, []>("aw_635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_635_cast_fp16 = einsum(equation = aw_635_equation_0, values = (var_4720_cast_fp16_13, var_4702_cast_fp16_13))[name = tensor<string, []>("aw_635_cast_fp16")];
            tensor<string, []> aw_637_equation_0 = const()[name = tensor<string, []>("aw_637_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_637_cast_fp16 = einsum(equation = aw_637_equation_0, values = (var_4720_cast_fp16_14, var_4702_cast_fp16_14))[name = tensor<string, []>("aw_637_cast_fp16")];
            tensor<string, []> aw_639_equation_0 = const()[name = tensor<string, []>("aw_639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_639_cast_fp16 = einsum(equation = aw_639_equation_0, values = (var_4720_cast_fp16_15, var_4702_cast_fp16_15))[name = tensor<string, []>("aw_639_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4786_cast_fp16 = softmax(axis = var_4650, x = aw_609_cast_fp16)[name = tensor<string, []>("op_4786_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4787_cast_fp16 = softmax(axis = var_4650, x = aw_611_cast_fp16)[name = tensor<string, []>("op_4787_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4788_cast_fp16 = softmax(axis = var_4650, x = aw_613_cast_fp16)[name = tensor<string, []>("op_4788_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4789_cast_fp16 = softmax(axis = var_4650, x = aw_615_cast_fp16)[name = tensor<string, []>("op_4789_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4790_cast_fp16 = softmax(axis = var_4650, x = aw_617_cast_fp16)[name = tensor<string, []>("op_4790_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4791_cast_fp16 = softmax(axis = var_4650, x = aw_619_cast_fp16)[name = tensor<string, []>("op_4791_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4792_cast_fp16 = softmax(axis = var_4650, x = aw_621_cast_fp16)[name = tensor<string, []>("op_4792_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4793_cast_fp16 = softmax(axis = var_4650, x = aw_623_cast_fp16)[name = tensor<string, []>("op_4793_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4794_cast_fp16 = softmax(axis = var_4650, x = aw_625_cast_fp16)[name = tensor<string, []>("op_4794_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4795_cast_fp16 = softmax(axis = var_4650, x = aw_627_cast_fp16)[name = tensor<string, []>("op_4795_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4796_cast_fp16 = softmax(axis = var_4650, x = aw_629_cast_fp16)[name = tensor<string, []>("op_4796_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4797_cast_fp16 = softmax(axis = var_4650, x = aw_631_cast_fp16)[name = tensor<string, []>("op_4797_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4798_cast_fp16 = softmax(axis = var_4650, x = aw_633_cast_fp16)[name = tensor<string, []>("op_4798_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4799_cast_fp16 = softmax(axis = var_4650, x = aw_635_cast_fp16)[name = tensor<string, []>("op_4799_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4800_cast_fp16 = softmax(axis = var_4650, x = aw_637_cast_fp16)[name = tensor<string, []>("op_4800_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_4801_cast_fp16 = softmax(axis = var_4650, x = aw_639_cast_fp16)[name = tensor<string, []>("op_4801_cast_fp16")];
            tensor<string, []> var_4803_equation_0 = const()[name = tensor<string, []>("op_4803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4803_cast_fp16 = einsum(equation = var_4803_equation_0, values = (var_4737_cast_fp16_0, var_4786_cast_fp16))[name = tensor<string, []>("op_4803_cast_fp16")];
            tensor<string, []> var_4805_equation_0 = const()[name = tensor<string, []>("op_4805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4805_cast_fp16 = einsum(equation = var_4805_equation_0, values = (var_4737_cast_fp16_1, var_4787_cast_fp16))[name = tensor<string, []>("op_4805_cast_fp16")];
            tensor<string, []> var_4807_equation_0 = const()[name = tensor<string, []>("op_4807_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4807_cast_fp16 = einsum(equation = var_4807_equation_0, values = (var_4737_cast_fp16_2, var_4788_cast_fp16))[name = tensor<string, []>("op_4807_cast_fp16")];
            tensor<string, []> var_4809_equation_0 = const()[name = tensor<string, []>("op_4809_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4809_cast_fp16 = einsum(equation = var_4809_equation_0, values = (var_4737_cast_fp16_3, var_4789_cast_fp16))[name = tensor<string, []>("op_4809_cast_fp16")];
            tensor<string, []> var_4811_equation_0 = const()[name = tensor<string, []>("op_4811_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4811_cast_fp16 = einsum(equation = var_4811_equation_0, values = (var_4737_cast_fp16_4, var_4790_cast_fp16))[name = tensor<string, []>("op_4811_cast_fp16")];
            tensor<string, []> var_4813_equation_0 = const()[name = tensor<string, []>("op_4813_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4813_cast_fp16 = einsum(equation = var_4813_equation_0, values = (var_4737_cast_fp16_5, var_4791_cast_fp16))[name = tensor<string, []>("op_4813_cast_fp16")];
            tensor<string, []> var_4815_equation_0 = const()[name = tensor<string, []>("op_4815_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4815_cast_fp16 = einsum(equation = var_4815_equation_0, values = (var_4737_cast_fp16_6, var_4792_cast_fp16))[name = tensor<string, []>("op_4815_cast_fp16")];
            tensor<string, []> var_4817_equation_0 = const()[name = tensor<string, []>("op_4817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4817_cast_fp16 = einsum(equation = var_4817_equation_0, values = (var_4737_cast_fp16_7, var_4793_cast_fp16))[name = tensor<string, []>("op_4817_cast_fp16")];
            tensor<string, []> var_4819_equation_0 = const()[name = tensor<string, []>("op_4819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4819_cast_fp16 = einsum(equation = var_4819_equation_0, values = (var_4737_cast_fp16_8, var_4794_cast_fp16))[name = tensor<string, []>("op_4819_cast_fp16")];
            tensor<string, []> var_4821_equation_0 = const()[name = tensor<string, []>("op_4821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4821_cast_fp16 = einsum(equation = var_4821_equation_0, values = (var_4737_cast_fp16_9, var_4795_cast_fp16))[name = tensor<string, []>("op_4821_cast_fp16")];
            tensor<string, []> var_4823_equation_0 = const()[name = tensor<string, []>("op_4823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4823_cast_fp16 = einsum(equation = var_4823_equation_0, values = (var_4737_cast_fp16_10, var_4796_cast_fp16))[name = tensor<string, []>("op_4823_cast_fp16")];
            tensor<string, []> var_4825_equation_0 = const()[name = tensor<string, []>("op_4825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4825_cast_fp16 = einsum(equation = var_4825_equation_0, values = (var_4737_cast_fp16_11, var_4797_cast_fp16))[name = tensor<string, []>("op_4825_cast_fp16")];
            tensor<string, []> var_4827_equation_0 = const()[name = tensor<string, []>("op_4827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4827_cast_fp16 = einsum(equation = var_4827_equation_0, values = (var_4737_cast_fp16_12, var_4798_cast_fp16))[name = tensor<string, []>("op_4827_cast_fp16")];
            tensor<string, []> var_4829_equation_0 = const()[name = tensor<string, []>("op_4829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4829_cast_fp16 = einsum(equation = var_4829_equation_0, values = (var_4737_cast_fp16_13, var_4799_cast_fp16))[name = tensor<string, []>("op_4829_cast_fp16")];
            tensor<string, []> var_4831_equation_0 = const()[name = tensor<string, []>("op_4831_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4737_cast_fp16_14, var_4800_cast_fp16))[name = tensor<string, []>("op_4831_cast_fp16")];
            tensor<string, []> var_4833_equation_0 = const()[name = tensor<string, []>("op_4833_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_4833_cast_fp16 = einsum(equation = var_4833_equation_0, values = (var_4737_cast_fp16_15, var_4801_cast_fp16))[name = tensor<string, []>("op_4833_cast_fp16")];
            tensor<bool, []> input_197_interleave_0 = const()[name = tensor<string, []>("input_197_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_197_cast_fp16 = concat(axis = var_4650, interleave = input_197_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16, var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16, var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16, var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = tensor<string, []>("input_197_cast_fp16")];
            tensor<string, []> var_4842_pad_type_0 = const()[name = tensor<string, []>("op_4842_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4842_strides_0 = const()[name = tensor<string, []>("op_4842_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4842_pad_0 = const()[name = tensor<string, []>("op_4842_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4842_dilations_0 = const()[name = tensor<string, []>("op_4842_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4842_groups_0 = const()[name = tensor<string, []>("op_4842_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_19_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123627136))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124151488))), name = tensor<string, []>("blocks_19_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_19_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124151616)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4842_cast_fp16 = conv(bias = blocks_19_attn_out_bias_to_fp16, dilations = var_4842_dilations_0, groups = var_4842_groups_0, pad = var_4842_pad_0, pad_type = var_4842_pad_type_0, strides = var_4842_strides_0, weight = blocks_19_attn_out_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor<string, []>("op_4842_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = var_4842_cast_fp16)[name = tensor<string, []>("inputs_79_cast_fp16")];
            tensor<int32, [1]> input_199_axes_0 = const()[name = tensor<string, []>("input_199_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_199_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_199_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124153728)))];
            tensor<fp16, [1024]> input_199_beta_0_to_fp16 = const()[name = tensor<string, []>("input_199_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124155840)))];
            tensor<fp16, []> var_4852_to_fp16 = const()[name = tensor<string, []>("op_4852_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_199_cast_fp16 = layer_norm(axes = input_199_axes_0, beta = input_199_beta_0_to_fp16, epsilon = var_4852_to_fp16, gamma = input_199_gamma_0_to_fp16, x = inputs_79_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<string, []> input_201_pad_type_0 = const()[name = tensor<string, []>("input_201_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_201_strides_0 = const()[name = tensor<string, []>("input_201_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_201_pad_0 = const()[name = tensor<string, []>("input_201_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_201_dilations_0 = const()[name = tensor<string, []>("input_201_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_201_groups_0 = const()[name = tensor<string, []>("input_201_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_19_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124157952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126255168))), name = tensor<string, []>("blocks_19_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_19_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126255296)))];
            tensor<fp16, [1, 4096, 1, 600]> input_201_cast_fp16 = conv(bias = blocks_19_mlp_0_bias_to_fp16, dilations = input_201_dilations_0, groups = input_201_groups_0, pad = input_201_pad_0, pad_type = input_201_pad_type_0, strides = input_201_strides_0, weight = blocks_19_mlp_0_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")];
            tensor<string, []> input_203_mode_0 = const()[name = tensor<string, []>("input_203_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_203_cast_fp16 = gelu(mode = input_203_mode_0, x = input_201_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
            tensor<string, []> var_4878_pad_type_0 = const()[name = tensor<string, []>("op_4878_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4878_strides_0 = const()[name = tensor<string, []>("op_4878_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4878_pad_0 = const()[name = tensor<string, []>("op_4878_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4878_dilations_0 = const()[name = tensor<string, []>("op_4878_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4878_groups_0 = const()[name = tensor<string, []>("op_4878_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_19_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126263552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128360768))), name = tensor<string, []>("blocks_19_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_19_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128360896)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4878_cast_fp16 = conv(bias = blocks_19_mlp_2_bias_to_fp16, dilations = var_4878_dilations_0, groups = var_4878_groups_0, pad = var_4878_pad_0, pad_type = var_4878_pad_type_0, strides = var_4878_strides_0, weight = blocks_19_mlp_2_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor<string, []>("op_4878_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = var_4878_cast_fp16)[name = tensor<string, []>("inputs_81_cast_fp16")];
            tensor<int32, []> var_4887 = const()[name = tensor<string, []>("op_4887"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_205_axes_0 = const()[name = tensor<string, []>("input_205_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_205_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_205_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128363008)))];
            tensor<fp16, [1024]> input_205_beta_0_to_fp16 = const()[name = tensor<string, []>("input_205_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128365120)))];
            tensor<fp16, []> var_4903_to_fp16 = const()[name = tensor<string, []>("op_4903_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_205_cast_fp16 = layer_norm(axes = input_205_axes_0, beta = input_205_beta_0_to_fp16, epsilon = var_4903_to_fp16, gamma = input_205_gamma_0_to_fp16, x = inputs_81_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")];
            tensor<string, []> q_41_pad_type_0 = const()[name = tensor<string, []>("q_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_41_strides_0 = const()[name = tensor<string, []>("q_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_41_pad_0 = const()[name = tensor<string, []>("q_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_41_dilations_0 = const()[name = tensor<string, []>("q_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_41_groups_0 = const()[name = tensor<string, []>("q_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_4938_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128367232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128891584))), name = tensor<string, []>("op_4938_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_4938_bias_0_to_fp16 = const()[name = tensor<string, []>("op_4938_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128891712)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4938_cast_fp16 = conv(bias = var_4938_bias_0_to_fp16, dilations = q_41_dilations_0, groups = q_41_groups_0, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = q_41_strides_0, weight = op_4938_weight_0_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor<string, []>("op_4938_cast_fp16")];
            tensor<string, []> k_41_pad_type_0 = const()[name = tensor<string, []>("k_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_41_strides_0 = const()[name = tensor<string, []>("k_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_41_pad_0 = const()[name = tensor<string, []>("k_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_41_dilations_0 = const()[name = tensor<string, []>("k_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_41_groups_0 = const()[name = tensor<string, []>("k_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_20_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128893824))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129418176))), name = tensor<string, []>("blocks_20_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_41_cast_fp16 = conv(dilations = k_41_dilations_0, groups = k_41_groups_0, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = k_41_strides_0, weight = blocks_20_attn_key_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor<string, []>("k_41_cast_fp16")];
            tensor<string, []> var_4936_pad_type_0 = const()[name = tensor<string, []>("op_4936_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_4936_strides_0 = const()[name = tensor<string, []>("op_4936_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4936_pad_0 = const()[name = tensor<string, []>("op_4936_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4936_dilations_0 = const()[name = tensor<string, []>("op_4936_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_4936_groups_0 = const()[name = tensor<string, []>("op_4936_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_20_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129418304))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129942656))), name = tensor<string, []>("blocks_20_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_20_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129942784)))];
            tensor<fp16, [1, 1024, 1, 600]> var_4936_cast_fp16 = conv(bias = blocks_20_attn_value_bias_to_fp16, dilations = var_4936_dilations_0, groups = var_4936_groups_0, pad = var_4936_pad_0, pad_type = var_4936_pad_type_0, strides = var_4936_strides_0, weight = blocks_20_attn_value_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor<string, []>("op_4936_cast_fp16")];
            tensor<int32, [16]> tile_60 = const()[name = tensor<string, []>("tile_60"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4939_axis_0 = const()[name = tensor<string, []>("op_4939_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4939_cast_fp16_15 = split(axis = var_4939_axis_0, split_sizes = tile_60, x = var_4938_cast_fp16)[name = tensor<string, []>("op_4939_cast_fp16")];
            tensor<int32, [4]> var_4956_perm_0 = const()[name = tensor<string, []>("op_4956_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_61 = const()[name = tensor<string, []>("tile_61"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4957_axis_0 = const()[name = tensor<string, []>("op_4957_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_4956_cast_fp16 = transpose(perm = var_4956_perm_0, x = k_41_cast_fp16)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_4957_cast_fp16_15 = split(axis = var_4957_axis_0, split_sizes = tile_61, x = var_4956_cast_fp16)[name = tensor<string, []>("op_4957_cast_fp16")];
            tensor<int32, [16]> tile_62 = const()[name = tensor<string, []>("tile_62"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_4974_axis_0 = const()[name = tensor<string, []>("op_4974_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_4974_cast_fp16_15 = split(axis = var_4974_axis_0, split_sizes = tile_62, x = var_4936_cast_fp16)[name = tensor<string, []>("op_4974_cast_fp16")];
            tensor<string, []> aw_641_equation_0 = const()[name = tensor<string, []>("aw_641_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_641_cast_fp16 = einsum(equation = aw_641_equation_0, values = (var_4957_cast_fp16_0, var_4939_cast_fp16_0))[name = tensor<string, []>("aw_641_cast_fp16")];
            tensor<string, []> aw_643_equation_0 = const()[name = tensor<string, []>("aw_643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_643_cast_fp16 = einsum(equation = aw_643_equation_0, values = (var_4957_cast_fp16_1, var_4939_cast_fp16_1))[name = tensor<string, []>("aw_643_cast_fp16")];
            tensor<string, []> aw_645_equation_0 = const()[name = tensor<string, []>("aw_645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_645_cast_fp16 = einsum(equation = aw_645_equation_0, values = (var_4957_cast_fp16_2, var_4939_cast_fp16_2))[name = tensor<string, []>("aw_645_cast_fp16")];
            tensor<string, []> aw_647_equation_0 = const()[name = tensor<string, []>("aw_647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_647_cast_fp16 = einsum(equation = aw_647_equation_0, values = (var_4957_cast_fp16_3, var_4939_cast_fp16_3))[name = tensor<string, []>("aw_647_cast_fp16")];
            tensor<string, []> aw_649_equation_0 = const()[name = tensor<string, []>("aw_649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_649_cast_fp16 = einsum(equation = aw_649_equation_0, values = (var_4957_cast_fp16_4, var_4939_cast_fp16_4))[name = tensor<string, []>("aw_649_cast_fp16")];
            tensor<string, []> aw_651_equation_0 = const()[name = tensor<string, []>("aw_651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_651_cast_fp16 = einsum(equation = aw_651_equation_0, values = (var_4957_cast_fp16_5, var_4939_cast_fp16_5))[name = tensor<string, []>("aw_651_cast_fp16")];
            tensor<string, []> aw_653_equation_0 = const()[name = tensor<string, []>("aw_653_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_653_cast_fp16 = einsum(equation = aw_653_equation_0, values = (var_4957_cast_fp16_6, var_4939_cast_fp16_6))[name = tensor<string, []>("aw_653_cast_fp16")];
            tensor<string, []> aw_655_equation_0 = const()[name = tensor<string, []>("aw_655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_655_cast_fp16 = einsum(equation = aw_655_equation_0, values = (var_4957_cast_fp16_7, var_4939_cast_fp16_7))[name = tensor<string, []>("aw_655_cast_fp16")];
            tensor<string, []> aw_657_equation_0 = const()[name = tensor<string, []>("aw_657_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_657_cast_fp16 = einsum(equation = aw_657_equation_0, values = (var_4957_cast_fp16_8, var_4939_cast_fp16_8))[name = tensor<string, []>("aw_657_cast_fp16")];
            tensor<string, []> aw_659_equation_0 = const()[name = tensor<string, []>("aw_659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_659_cast_fp16 = einsum(equation = aw_659_equation_0, values = (var_4957_cast_fp16_9, var_4939_cast_fp16_9))[name = tensor<string, []>("aw_659_cast_fp16")];
            tensor<string, []> aw_661_equation_0 = const()[name = tensor<string, []>("aw_661_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_661_cast_fp16 = einsum(equation = aw_661_equation_0, values = (var_4957_cast_fp16_10, var_4939_cast_fp16_10))[name = tensor<string, []>("aw_661_cast_fp16")];
            tensor<string, []> aw_663_equation_0 = const()[name = tensor<string, []>("aw_663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_663_cast_fp16 = einsum(equation = aw_663_equation_0, values = (var_4957_cast_fp16_11, var_4939_cast_fp16_11))[name = tensor<string, []>("aw_663_cast_fp16")];
            tensor<string, []> aw_665_equation_0 = const()[name = tensor<string, []>("aw_665_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_665_cast_fp16 = einsum(equation = aw_665_equation_0, values = (var_4957_cast_fp16_12, var_4939_cast_fp16_12))[name = tensor<string, []>("aw_665_cast_fp16")];
            tensor<string, []> aw_667_equation_0 = const()[name = tensor<string, []>("aw_667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_667_cast_fp16 = einsum(equation = aw_667_equation_0, values = (var_4957_cast_fp16_13, var_4939_cast_fp16_13))[name = tensor<string, []>("aw_667_cast_fp16")];
            tensor<string, []> aw_669_equation_0 = const()[name = tensor<string, []>("aw_669_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_669_cast_fp16 = einsum(equation = aw_669_equation_0, values = (var_4957_cast_fp16_14, var_4939_cast_fp16_14))[name = tensor<string, []>("aw_669_cast_fp16")];
            tensor<string, []> aw_671_equation_0 = const()[name = tensor<string, []>("aw_671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_671_cast_fp16 = einsum(equation = aw_671_equation_0, values = (var_4957_cast_fp16_15, var_4939_cast_fp16_15))[name = tensor<string, []>("aw_671_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5023_cast_fp16 = softmax(axis = var_4887, x = aw_641_cast_fp16)[name = tensor<string, []>("op_5023_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5024_cast_fp16 = softmax(axis = var_4887, x = aw_643_cast_fp16)[name = tensor<string, []>("op_5024_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5025_cast_fp16 = softmax(axis = var_4887, x = aw_645_cast_fp16)[name = tensor<string, []>("op_5025_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5026_cast_fp16 = softmax(axis = var_4887, x = aw_647_cast_fp16)[name = tensor<string, []>("op_5026_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5027_cast_fp16 = softmax(axis = var_4887, x = aw_649_cast_fp16)[name = tensor<string, []>("op_5027_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5028_cast_fp16 = softmax(axis = var_4887, x = aw_651_cast_fp16)[name = tensor<string, []>("op_5028_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5029_cast_fp16 = softmax(axis = var_4887, x = aw_653_cast_fp16)[name = tensor<string, []>("op_5029_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5030_cast_fp16 = softmax(axis = var_4887, x = aw_655_cast_fp16)[name = tensor<string, []>("op_5030_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5031_cast_fp16 = softmax(axis = var_4887, x = aw_657_cast_fp16)[name = tensor<string, []>("op_5031_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5032_cast_fp16 = softmax(axis = var_4887, x = aw_659_cast_fp16)[name = tensor<string, []>("op_5032_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5033_cast_fp16 = softmax(axis = var_4887, x = aw_661_cast_fp16)[name = tensor<string, []>("op_5033_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5034_cast_fp16 = softmax(axis = var_4887, x = aw_663_cast_fp16)[name = tensor<string, []>("op_5034_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5035_cast_fp16 = softmax(axis = var_4887, x = aw_665_cast_fp16)[name = tensor<string, []>("op_5035_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5036_cast_fp16 = softmax(axis = var_4887, x = aw_667_cast_fp16)[name = tensor<string, []>("op_5036_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5037_cast_fp16 = softmax(axis = var_4887, x = aw_669_cast_fp16)[name = tensor<string, []>("op_5037_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5038_cast_fp16 = softmax(axis = var_4887, x = aw_671_cast_fp16)[name = tensor<string, []>("op_5038_cast_fp16")];
            tensor<string, []> var_5040_equation_0 = const()[name = tensor<string, []>("op_5040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5040_cast_fp16 = einsum(equation = var_5040_equation_0, values = (var_4974_cast_fp16_0, var_5023_cast_fp16))[name = tensor<string, []>("op_5040_cast_fp16")];
            tensor<string, []> var_5042_equation_0 = const()[name = tensor<string, []>("op_5042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5042_cast_fp16 = einsum(equation = var_5042_equation_0, values = (var_4974_cast_fp16_1, var_5024_cast_fp16))[name = tensor<string, []>("op_5042_cast_fp16")];
            tensor<string, []> var_5044_equation_0 = const()[name = tensor<string, []>("op_5044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5044_cast_fp16 = einsum(equation = var_5044_equation_0, values = (var_4974_cast_fp16_2, var_5025_cast_fp16))[name = tensor<string, []>("op_5044_cast_fp16")];
            tensor<string, []> var_5046_equation_0 = const()[name = tensor<string, []>("op_5046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5046_cast_fp16 = einsum(equation = var_5046_equation_0, values = (var_4974_cast_fp16_3, var_5026_cast_fp16))[name = tensor<string, []>("op_5046_cast_fp16")];
            tensor<string, []> var_5048_equation_0 = const()[name = tensor<string, []>("op_5048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5048_cast_fp16 = einsum(equation = var_5048_equation_0, values = (var_4974_cast_fp16_4, var_5027_cast_fp16))[name = tensor<string, []>("op_5048_cast_fp16")];
            tensor<string, []> var_5050_equation_0 = const()[name = tensor<string, []>("op_5050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5050_cast_fp16 = einsum(equation = var_5050_equation_0, values = (var_4974_cast_fp16_5, var_5028_cast_fp16))[name = tensor<string, []>("op_5050_cast_fp16")];
            tensor<string, []> var_5052_equation_0 = const()[name = tensor<string, []>("op_5052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5052_cast_fp16 = einsum(equation = var_5052_equation_0, values = (var_4974_cast_fp16_6, var_5029_cast_fp16))[name = tensor<string, []>("op_5052_cast_fp16")];
            tensor<string, []> var_5054_equation_0 = const()[name = tensor<string, []>("op_5054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5054_cast_fp16 = einsum(equation = var_5054_equation_0, values = (var_4974_cast_fp16_7, var_5030_cast_fp16))[name = tensor<string, []>("op_5054_cast_fp16")];
            tensor<string, []> var_5056_equation_0 = const()[name = tensor<string, []>("op_5056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5056_cast_fp16 = einsum(equation = var_5056_equation_0, values = (var_4974_cast_fp16_8, var_5031_cast_fp16))[name = tensor<string, []>("op_5056_cast_fp16")];
            tensor<string, []> var_5058_equation_0 = const()[name = tensor<string, []>("op_5058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5058_cast_fp16 = einsum(equation = var_5058_equation_0, values = (var_4974_cast_fp16_9, var_5032_cast_fp16))[name = tensor<string, []>("op_5058_cast_fp16")];
            tensor<string, []> var_5060_equation_0 = const()[name = tensor<string, []>("op_5060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5060_cast_fp16 = einsum(equation = var_5060_equation_0, values = (var_4974_cast_fp16_10, var_5033_cast_fp16))[name = tensor<string, []>("op_5060_cast_fp16")];
            tensor<string, []> var_5062_equation_0 = const()[name = tensor<string, []>("op_5062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5062_cast_fp16 = einsum(equation = var_5062_equation_0, values = (var_4974_cast_fp16_11, var_5034_cast_fp16))[name = tensor<string, []>("op_5062_cast_fp16")];
            tensor<string, []> var_5064_equation_0 = const()[name = tensor<string, []>("op_5064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5064_cast_fp16 = einsum(equation = var_5064_equation_0, values = (var_4974_cast_fp16_12, var_5035_cast_fp16))[name = tensor<string, []>("op_5064_cast_fp16")];
            tensor<string, []> var_5066_equation_0 = const()[name = tensor<string, []>("op_5066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5066_cast_fp16 = einsum(equation = var_5066_equation_0, values = (var_4974_cast_fp16_13, var_5036_cast_fp16))[name = tensor<string, []>("op_5066_cast_fp16")];
            tensor<string, []> var_5068_equation_0 = const()[name = tensor<string, []>("op_5068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5068_cast_fp16 = einsum(equation = var_5068_equation_0, values = (var_4974_cast_fp16_14, var_5037_cast_fp16))[name = tensor<string, []>("op_5068_cast_fp16")];
            tensor<string, []> var_5070_equation_0 = const()[name = tensor<string, []>("op_5070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5070_cast_fp16 = einsum(equation = var_5070_equation_0, values = (var_4974_cast_fp16_15, var_5038_cast_fp16))[name = tensor<string, []>("op_5070_cast_fp16")];
            tensor<bool, []> input_207_interleave_0 = const()[name = tensor<string, []>("input_207_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_207_cast_fp16 = concat(axis = var_4887, interleave = input_207_interleave_0, values = (var_5040_cast_fp16, var_5042_cast_fp16, var_5044_cast_fp16, var_5046_cast_fp16, var_5048_cast_fp16, var_5050_cast_fp16, var_5052_cast_fp16, var_5054_cast_fp16, var_5056_cast_fp16, var_5058_cast_fp16, var_5060_cast_fp16, var_5062_cast_fp16, var_5064_cast_fp16, var_5066_cast_fp16, var_5068_cast_fp16, var_5070_cast_fp16))[name = tensor<string, []>("input_207_cast_fp16")];
            tensor<string, []> var_5079_pad_type_0 = const()[name = tensor<string, []>("op_5079_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5079_strides_0 = const()[name = tensor<string, []>("op_5079_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5079_pad_0 = const()[name = tensor<string, []>("op_5079_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5079_dilations_0 = const()[name = tensor<string, []>("op_5079_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5079_groups_0 = const()[name = tensor<string, []>("op_5079_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_20_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129944896))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130469248))), name = tensor<string, []>("blocks_20_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_20_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130469376)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5079_cast_fp16 = conv(bias = blocks_20_attn_out_bias_to_fp16, dilations = var_5079_dilations_0, groups = var_5079_groups_0, pad = var_5079_pad_0, pad_type = var_5079_pad_type_0, strides = var_5079_strides_0, weight = blocks_20_attn_out_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor<string, []>("op_5079_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = var_5079_cast_fp16)[name = tensor<string, []>("inputs_83_cast_fp16")];
            tensor<int32, [1]> input_209_axes_0 = const()[name = tensor<string, []>("input_209_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_209_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_209_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130471488)))];
            tensor<fp16, [1024]> input_209_beta_0_to_fp16 = const()[name = tensor<string, []>("input_209_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130473600)))];
            tensor<fp16, []> var_5089_to_fp16 = const()[name = tensor<string, []>("op_5089_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_209_cast_fp16 = layer_norm(axes = input_209_axes_0, beta = input_209_beta_0_to_fp16, epsilon = var_5089_to_fp16, gamma = input_209_gamma_0_to_fp16, x = inputs_83_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
            tensor<string, []> input_211_pad_type_0 = const()[name = tensor<string, []>("input_211_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_211_strides_0 = const()[name = tensor<string, []>("input_211_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_211_pad_0 = const()[name = tensor<string, []>("input_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_211_dilations_0 = const()[name = tensor<string, []>("input_211_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_211_groups_0 = const()[name = tensor<string, []>("input_211_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_20_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130475712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132572928))), name = tensor<string, []>("blocks_20_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_20_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132573056)))];
            tensor<fp16, [1, 4096, 1, 600]> input_211_cast_fp16 = conv(bias = blocks_20_mlp_0_bias_to_fp16, dilations = input_211_dilations_0, groups = input_211_groups_0, pad = input_211_pad_0, pad_type = input_211_pad_type_0, strides = input_211_strides_0, weight = blocks_20_mlp_0_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")];
            tensor<string, []> input_213_mode_0 = const()[name = tensor<string, []>("input_213_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_213_cast_fp16 = gelu(mode = input_213_mode_0, x = input_211_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")];
            tensor<string, []> var_5115_pad_type_0 = const()[name = tensor<string, []>("op_5115_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5115_strides_0 = const()[name = tensor<string, []>("op_5115_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5115_pad_0 = const()[name = tensor<string, []>("op_5115_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5115_dilations_0 = const()[name = tensor<string, []>("op_5115_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5115_groups_0 = const()[name = tensor<string, []>("op_5115_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_20_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132581312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134678528))), name = tensor<string, []>("blocks_20_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_20_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134678656)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5115_cast_fp16 = conv(bias = blocks_20_mlp_2_bias_to_fp16, dilations = var_5115_dilations_0, groups = var_5115_groups_0, pad = var_5115_pad_0, pad_type = var_5115_pad_type_0, strides = var_5115_strides_0, weight = blocks_20_mlp_2_weight_to_fp16_palettized, x = input_213_cast_fp16)[name = tensor<string, []>("op_5115_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = var_5115_cast_fp16)[name = tensor<string, []>("inputs_85_cast_fp16")];
            tensor<int32, []> var_5124 = const()[name = tensor<string, []>("op_5124"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_215_axes_0 = const()[name = tensor<string, []>("input_215_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_215_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_215_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134680768)))];
            tensor<fp16, [1024]> input_215_beta_0_to_fp16 = const()[name = tensor<string, []>("input_215_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134682880)))];
            tensor<fp16, []> var_5140_to_fp16 = const()[name = tensor<string, []>("op_5140_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = input_215_beta_0_to_fp16, epsilon = var_5140_to_fp16, gamma = input_215_gamma_0_to_fp16, x = inputs_85_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<string, []> q_43_pad_type_0 = const()[name = tensor<string, []>("q_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_43_strides_0 = const()[name = tensor<string, []>("q_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_43_pad_0 = const()[name = tensor<string, []>("q_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_43_dilations_0 = const()[name = tensor<string, []>("q_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_43_groups_0 = const()[name = tensor<string, []>("q_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_5175_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134684992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135209344))), name = tensor<string, []>("op_5175_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_5175_bias_0_to_fp16 = const()[name = tensor<string, []>("op_5175_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135209472)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5175_cast_fp16 = conv(bias = var_5175_bias_0_to_fp16, dilations = q_43_dilations_0, groups = q_43_groups_0, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = q_43_strides_0, weight = op_5175_weight_0_to_fp16_palettized, x = input_215_cast_fp16)[name = tensor<string, []>("op_5175_cast_fp16")];
            tensor<string, []> k_43_pad_type_0 = const()[name = tensor<string, []>("k_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_43_strides_0 = const()[name = tensor<string, []>("k_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_43_pad_0 = const()[name = tensor<string, []>("k_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_43_dilations_0 = const()[name = tensor<string, []>("k_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_43_groups_0 = const()[name = tensor<string, []>("k_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_21_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135211584))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135735936))), name = tensor<string, []>("blocks_21_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_43_cast_fp16 = conv(dilations = k_43_dilations_0, groups = k_43_groups_0, pad = k_43_pad_0, pad_type = k_43_pad_type_0, strides = k_43_strides_0, weight = blocks_21_attn_key_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = tensor<string, []>("k_43_cast_fp16")];
            tensor<string, []> var_5173_pad_type_0 = const()[name = tensor<string, []>("op_5173_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5173_strides_0 = const()[name = tensor<string, []>("op_5173_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5173_pad_0 = const()[name = tensor<string, []>("op_5173_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5173_dilations_0 = const()[name = tensor<string, []>("op_5173_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5173_groups_0 = const()[name = tensor<string, []>("op_5173_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_21_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135736064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136260416))), name = tensor<string, []>("blocks_21_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_21_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136260544)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5173_cast_fp16 = conv(bias = blocks_21_attn_value_bias_to_fp16, dilations = var_5173_dilations_0, groups = var_5173_groups_0, pad = var_5173_pad_0, pad_type = var_5173_pad_type_0, strides = var_5173_strides_0, weight = blocks_21_attn_value_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = tensor<string, []>("op_5173_cast_fp16")];
            tensor<int32, [16]> tile_63 = const()[name = tensor<string, []>("tile_63"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5176_axis_0 = const()[name = tensor<string, []>("op_5176_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5176_cast_fp16_15 = split(axis = var_5176_axis_0, split_sizes = tile_63, x = var_5175_cast_fp16)[name = tensor<string, []>("op_5176_cast_fp16")];
            tensor<int32, [4]> var_5193_perm_0 = const()[name = tensor<string, []>("op_5193_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_64 = const()[name = tensor<string, []>("tile_64"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5194_axis_0 = const()[name = tensor<string, []>("op_5194_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_5193_cast_fp16 = transpose(perm = var_5193_perm_0, x = k_43_cast_fp16)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_5194_cast_fp16_15 = split(axis = var_5194_axis_0, split_sizes = tile_64, x = var_5193_cast_fp16)[name = tensor<string, []>("op_5194_cast_fp16")];
            tensor<int32, [16]> tile_65 = const()[name = tensor<string, []>("tile_65"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5211_axis_0 = const()[name = tensor<string, []>("op_5211_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5211_cast_fp16_15 = split(axis = var_5211_axis_0, split_sizes = tile_65, x = var_5173_cast_fp16)[name = tensor<string, []>("op_5211_cast_fp16")];
            tensor<string, []> aw_673_equation_0 = const()[name = tensor<string, []>("aw_673_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_673_cast_fp16 = einsum(equation = aw_673_equation_0, values = (var_5194_cast_fp16_0, var_5176_cast_fp16_0))[name = tensor<string, []>("aw_673_cast_fp16")];
            tensor<string, []> aw_675_equation_0 = const()[name = tensor<string, []>("aw_675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_675_cast_fp16 = einsum(equation = aw_675_equation_0, values = (var_5194_cast_fp16_1, var_5176_cast_fp16_1))[name = tensor<string, []>("aw_675_cast_fp16")];
            tensor<string, []> aw_677_equation_0 = const()[name = tensor<string, []>("aw_677_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_677_cast_fp16 = einsum(equation = aw_677_equation_0, values = (var_5194_cast_fp16_2, var_5176_cast_fp16_2))[name = tensor<string, []>("aw_677_cast_fp16")];
            tensor<string, []> aw_679_equation_0 = const()[name = tensor<string, []>("aw_679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_679_cast_fp16 = einsum(equation = aw_679_equation_0, values = (var_5194_cast_fp16_3, var_5176_cast_fp16_3))[name = tensor<string, []>("aw_679_cast_fp16")];
            tensor<string, []> aw_681_equation_0 = const()[name = tensor<string, []>("aw_681_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_681_cast_fp16 = einsum(equation = aw_681_equation_0, values = (var_5194_cast_fp16_4, var_5176_cast_fp16_4))[name = tensor<string, []>("aw_681_cast_fp16")];
            tensor<string, []> aw_683_equation_0 = const()[name = tensor<string, []>("aw_683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_683_cast_fp16 = einsum(equation = aw_683_equation_0, values = (var_5194_cast_fp16_5, var_5176_cast_fp16_5))[name = tensor<string, []>("aw_683_cast_fp16")];
            tensor<string, []> aw_685_equation_0 = const()[name = tensor<string, []>("aw_685_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_685_cast_fp16 = einsum(equation = aw_685_equation_0, values = (var_5194_cast_fp16_6, var_5176_cast_fp16_6))[name = tensor<string, []>("aw_685_cast_fp16")];
            tensor<string, []> aw_687_equation_0 = const()[name = tensor<string, []>("aw_687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_687_cast_fp16 = einsum(equation = aw_687_equation_0, values = (var_5194_cast_fp16_7, var_5176_cast_fp16_7))[name = tensor<string, []>("aw_687_cast_fp16")];
            tensor<string, []> aw_689_equation_0 = const()[name = tensor<string, []>("aw_689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_689_cast_fp16 = einsum(equation = aw_689_equation_0, values = (var_5194_cast_fp16_8, var_5176_cast_fp16_8))[name = tensor<string, []>("aw_689_cast_fp16")];
            tensor<string, []> aw_691_equation_0 = const()[name = tensor<string, []>("aw_691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_691_cast_fp16 = einsum(equation = aw_691_equation_0, values = (var_5194_cast_fp16_9, var_5176_cast_fp16_9))[name = tensor<string, []>("aw_691_cast_fp16")];
            tensor<string, []> aw_693_equation_0 = const()[name = tensor<string, []>("aw_693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_693_cast_fp16 = einsum(equation = aw_693_equation_0, values = (var_5194_cast_fp16_10, var_5176_cast_fp16_10))[name = tensor<string, []>("aw_693_cast_fp16")];
            tensor<string, []> aw_695_equation_0 = const()[name = tensor<string, []>("aw_695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_695_cast_fp16 = einsum(equation = aw_695_equation_0, values = (var_5194_cast_fp16_11, var_5176_cast_fp16_11))[name = tensor<string, []>("aw_695_cast_fp16")];
            tensor<string, []> aw_697_equation_0 = const()[name = tensor<string, []>("aw_697_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_697_cast_fp16 = einsum(equation = aw_697_equation_0, values = (var_5194_cast_fp16_12, var_5176_cast_fp16_12))[name = tensor<string, []>("aw_697_cast_fp16")];
            tensor<string, []> aw_699_equation_0 = const()[name = tensor<string, []>("aw_699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_699_cast_fp16 = einsum(equation = aw_699_equation_0, values = (var_5194_cast_fp16_13, var_5176_cast_fp16_13))[name = tensor<string, []>("aw_699_cast_fp16")];
            tensor<string, []> aw_701_equation_0 = const()[name = tensor<string, []>("aw_701_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_701_cast_fp16 = einsum(equation = aw_701_equation_0, values = (var_5194_cast_fp16_14, var_5176_cast_fp16_14))[name = tensor<string, []>("aw_701_cast_fp16")];
            tensor<string, []> aw_703_equation_0 = const()[name = tensor<string, []>("aw_703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_703_cast_fp16 = einsum(equation = aw_703_equation_0, values = (var_5194_cast_fp16_15, var_5176_cast_fp16_15))[name = tensor<string, []>("aw_703_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5260_cast_fp16 = softmax(axis = var_5124, x = aw_673_cast_fp16)[name = tensor<string, []>("op_5260_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5261_cast_fp16 = softmax(axis = var_5124, x = aw_675_cast_fp16)[name = tensor<string, []>("op_5261_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5262_cast_fp16 = softmax(axis = var_5124, x = aw_677_cast_fp16)[name = tensor<string, []>("op_5262_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5263_cast_fp16 = softmax(axis = var_5124, x = aw_679_cast_fp16)[name = tensor<string, []>("op_5263_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5264_cast_fp16 = softmax(axis = var_5124, x = aw_681_cast_fp16)[name = tensor<string, []>("op_5264_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5265_cast_fp16 = softmax(axis = var_5124, x = aw_683_cast_fp16)[name = tensor<string, []>("op_5265_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5266_cast_fp16 = softmax(axis = var_5124, x = aw_685_cast_fp16)[name = tensor<string, []>("op_5266_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5267_cast_fp16 = softmax(axis = var_5124, x = aw_687_cast_fp16)[name = tensor<string, []>("op_5267_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5268_cast_fp16 = softmax(axis = var_5124, x = aw_689_cast_fp16)[name = tensor<string, []>("op_5268_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5269_cast_fp16 = softmax(axis = var_5124, x = aw_691_cast_fp16)[name = tensor<string, []>("op_5269_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5270_cast_fp16 = softmax(axis = var_5124, x = aw_693_cast_fp16)[name = tensor<string, []>("op_5270_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5271_cast_fp16 = softmax(axis = var_5124, x = aw_695_cast_fp16)[name = tensor<string, []>("op_5271_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5272_cast_fp16 = softmax(axis = var_5124, x = aw_697_cast_fp16)[name = tensor<string, []>("op_5272_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5273_cast_fp16 = softmax(axis = var_5124, x = aw_699_cast_fp16)[name = tensor<string, []>("op_5273_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5274_cast_fp16 = softmax(axis = var_5124, x = aw_701_cast_fp16)[name = tensor<string, []>("op_5274_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5275_cast_fp16 = softmax(axis = var_5124, x = aw_703_cast_fp16)[name = tensor<string, []>("op_5275_cast_fp16")];
            tensor<string, []> var_5277_equation_0 = const()[name = tensor<string, []>("op_5277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5277_cast_fp16 = einsum(equation = var_5277_equation_0, values = (var_5211_cast_fp16_0, var_5260_cast_fp16))[name = tensor<string, []>("op_5277_cast_fp16")];
            tensor<string, []> var_5279_equation_0 = const()[name = tensor<string, []>("op_5279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5279_cast_fp16 = einsum(equation = var_5279_equation_0, values = (var_5211_cast_fp16_1, var_5261_cast_fp16))[name = tensor<string, []>("op_5279_cast_fp16")];
            tensor<string, []> var_5281_equation_0 = const()[name = tensor<string, []>("op_5281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5281_cast_fp16 = einsum(equation = var_5281_equation_0, values = (var_5211_cast_fp16_2, var_5262_cast_fp16))[name = tensor<string, []>("op_5281_cast_fp16")];
            tensor<string, []> var_5283_equation_0 = const()[name = tensor<string, []>("op_5283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5283_cast_fp16 = einsum(equation = var_5283_equation_0, values = (var_5211_cast_fp16_3, var_5263_cast_fp16))[name = tensor<string, []>("op_5283_cast_fp16")];
            tensor<string, []> var_5285_equation_0 = const()[name = tensor<string, []>("op_5285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5285_cast_fp16 = einsum(equation = var_5285_equation_0, values = (var_5211_cast_fp16_4, var_5264_cast_fp16))[name = tensor<string, []>("op_5285_cast_fp16")];
            tensor<string, []> var_5287_equation_0 = const()[name = tensor<string, []>("op_5287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5287_cast_fp16 = einsum(equation = var_5287_equation_0, values = (var_5211_cast_fp16_5, var_5265_cast_fp16))[name = tensor<string, []>("op_5287_cast_fp16")];
            tensor<string, []> var_5289_equation_0 = const()[name = tensor<string, []>("op_5289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5289_cast_fp16 = einsum(equation = var_5289_equation_0, values = (var_5211_cast_fp16_6, var_5266_cast_fp16))[name = tensor<string, []>("op_5289_cast_fp16")];
            tensor<string, []> var_5291_equation_0 = const()[name = tensor<string, []>("op_5291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5291_cast_fp16 = einsum(equation = var_5291_equation_0, values = (var_5211_cast_fp16_7, var_5267_cast_fp16))[name = tensor<string, []>("op_5291_cast_fp16")];
            tensor<string, []> var_5293_equation_0 = const()[name = tensor<string, []>("op_5293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5293_cast_fp16 = einsum(equation = var_5293_equation_0, values = (var_5211_cast_fp16_8, var_5268_cast_fp16))[name = tensor<string, []>("op_5293_cast_fp16")];
            tensor<string, []> var_5295_equation_0 = const()[name = tensor<string, []>("op_5295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5295_cast_fp16 = einsum(equation = var_5295_equation_0, values = (var_5211_cast_fp16_9, var_5269_cast_fp16))[name = tensor<string, []>("op_5295_cast_fp16")];
            tensor<string, []> var_5297_equation_0 = const()[name = tensor<string, []>("op_5297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5297_cast_fp16 = einsum(equation = var_5297_equation_0, values = (var_5211_cast_fp16_10, var_5270_cast_fp16))[name = tensor<string, []>("op_5297_cast_fp16")];
            tensor<string, []> var_5299_equation_0 = const()[name = tensor<string, []>("op_5299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5299_cast_fp16 = einsum(equation = var_5299_equation_0, values = (var_5211_cast_fp16_11, var_5271_cast_fp16))[name = tensor<string, []>("op_5299_cast_fp16")];
            tensor<string, []> var_5301_equation_0 = const()[name = tensor<string, []>("op_5301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5301_cast_fp16 = einsum(equation = var_5301_equation_0, values = (var_5211_cast_fp16_12, var_5272_cast_fp16))[name = tensor<string, []>("op_5301_cast_fp16")];
            tensor<string, []> var_5303_equation_0 = const()[name = tensor<string, []>("op_5303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5303_cast_fp16 = einsum(equation = var_5303_equation_0, values = (var_5211_cast_fp16_13, var_5273_cast_fp16))[name = tensor<string, []>("op_5303_cast_fp16")];
            tensor<string, []> var_5305_equation_0 = const()[name = tensor<string, []>("op_5305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5305_cast_fp16 = einsum(equation = var_5305_equation_0, values = (var_5211_cast_fp16_14, var_5274_cast_fp16))[name = tensor<string, []>("op_5305_cast_fp16")];
            tensor<string, []> var_5307_equation_0 = const()[name = tensor<string, []>("op_5307_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5307_cast_fp16 = einsum(equation = var_5307_equation_0, values = (var_5211_cast_fp16_15, var_5275_cast_fp16))[name = tensor<string, []>("op_5307_cast_fp16")];
            tensor<bool, []> input_217_interleave_0 = const()[name = tensor<string, []>("input_217_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_217_cast_fp16 = concat(axis = var_5124, interleave = input_217_interleave_0, values = (var_5277_cast_fp16, var_5279_cast_fp16, var_5281_cast_fp16, var_5283_cast_fp16, var_5285_cast_fp16, var_5287_cast_fp16, var_5289_cast_fp16, var_5291_cast_fp16, var_5293_cast_fp16, var_5295_cast_fp16, var_5297_cast_fp16, var_5299_cast_fp16, var_5301_cast_fp16, var_5303_cast_fp16, var_5305_cast_fp16, var_5307_cast_fp16))[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<string, []> var_5316_pad_type_0 = const()[name = tensor<string, []>("op_5316_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5316_strides_0 = const()[name = tensor<string, []>("op_5316_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5316_pad_0 = const()[name = tensor<string, []>("op_5316_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5316_dilations_0 = const()[name = tensor<string, []>("op_5316_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5316_groups_0 = const()[name = tensor<string, []>("op_5316_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_21_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136262656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136787008))), name = tensor<string, []>("blocks_21_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_21_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136787136)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5316_cast_fp16 = conv(bias = blocks_21_attn_out_bias_to_fp16, dilations = var_5316_dilations_0, groups = var_5316_groups_0, pad = var_5316_pad_0, pad_type = var_5316_pad_type_0, strides = var_5316_strides_0, weight = blocks_21_attn_out_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor<string, []>("op_5316_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = var_5316_cast_fp16)[name = tensor<string, []>("inputs_87_cast_fp16")];
            tensor<int32, [1]> input_219_axes_0 = const()[name = tensor<string, []>("input_219_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_219_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_219_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136789248)))];
            tensor<fp16, [1024]> input_219_beta_0_to_fp16 = const()[name = tensor<string, []>("input_219_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136791360)))];
            tensor<fp16, []> var_5326_to_fp16 = const()[name = tensor<string, []>("op_5326_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_219_cast_fp16 = layer_norm(axes = input_219_axes_0, beta = input_219_beta_0_to_fp16, epsilon = var_5326_to_fp16, gamma = input_219_gamma_0_to_fp16, x = inputs_87_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")];
            tensor<string, []> input_221_pad_type_0 = const()[name = tensor<string, []>("input_221_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_221_strides_0 = const()[name = tensor<string, []>("input_221_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_221_pad_0 = const()[name = tensor<string, []>("input_221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_221_dilations_0 = const()[name = tensor<string, []>("input_221_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_221_groups_0 = const()[name = tensor<string, []>("input_221_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_21_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136793472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138890688))), name = tensor<string, []>("blocks_21_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_21_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138890816)))];
            tensor<fp16, [1, 4096, 1, 600]> input_221_cast_fp16 = conv(bias = blocks_21_mlp_0_bias_to_fp16, dilations = input_221_dilations_0, groups = input_221_groups_0, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = input_221_strides_0, weight = blocks_21_mlp_0_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")];
            tensor<string, []> input_223_mode_0 = const()[name = tensor<string, []>("input_223_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = input_221_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")];
            tensor<string, []> var_5352_pad_type_0 = const()[name = tensor<string, []>("op_5352_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5352_strides_0 = const()[name = tensor<string, []>("op_5352_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5352_pad_0 = const()[name = tensor<string, []>("op_5352_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5352_dilations_0 = const()[name = tensor<string, []>("op_5352_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5352_groups_0 = const()[name = tensor<string, []>("op_5352_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_21_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138899072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140996288))), name = tensor<string, []>("blocks_21_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_21_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140996416)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5352_cast_fp16 = conv(bias = blocks_21_mlp_2_bias_to_fp16, dilations = var_5352_dilations_0, groups = var_5352_groups_0, pad = var_5352_pad_0, pad_type = var_5352_pad_type_0, strides = var_5352_strides_0, weight = blocks_21_mlp_2_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor<string, []>("op_5352_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = var_5352_cast_fp16)[name = tensor<string, []>("inputs_89_cast_fp16")];
            tensor<int32, []> var_5361 = const()[name = tensor<string, []>("op_5361"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_225_axes_0 = const()[name = tensor<string, []>("input_225_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_225_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_225_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140998528)))];
            tensor<fp16, [1024]> input_225_beta_0_to_fp16 = const()[name = tensor<string, []>("input_225_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141000640)))];
            tensor<fp16, []> var_5377_to_fp16 = const()[name = tensor<string, []>("op_5377_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_225_cast_fp16 = layer_norm(axes = input_225_axes_0, beta = input_225_beta_0_to_fp16, epsilon = var_5377_to_fp16, gamma = input_225_gamma_0_to_fp16, x = inputs_89_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
            tensor<string, []> q_45_pad_type_0 = const()[name = tensor<string, []>("q_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_45_strides_0 = const()[name = tensor<string, []>("q_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_45_pad_0 = const()[name = tensor<string, []>("q_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_45_dilations_0 = const()[name = tensor<string, []>("q_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_45_groups_0 = const()[name = tensor<string, []>("q_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_5412_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141002752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141527104))), name = tensor<string, []>("op_5412_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_5412_bias_0_to_fp16 = const()[name = tensor<string, []>("op_5412_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141527232)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5412_cast_fp16 = conv(bias = var_5412_bias_0_to_fp16, dilations = q_45_dilations_0, groups = q_45_groups_0, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = q_45_strides_0, weight = op_5412_weight_0_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor<string, []>("op_5412_cast_fp16")];
            tensor<string, []> k_45_pad_type_0 = const()[name = tensor<string, []>("k_45_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_45_strides_0 = const()[name = tensor<string, []>("k_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_45_pad_0 = const()[name = tensor<string, []>("k_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_45_dilations_0 = const()[name = tensor<string, []>("k_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_45_groups_0 = const()[name = tensor<string, []>("k_45_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_22_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141529344))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142053696))), name = tensor<string, []>("blocks_22_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_45_cast_fp16 = conv(dilations = k_45_dilations_0, groups = k_45_groups_0, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = k_45_strides_0, weight = blocks_22_attn_key_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor<string, []>("k_45_cast_fp16")];
            tensor<string, []> var_5410_pad_type_0 = const()[name = tensor<string, []>("op_5410_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5410_strides_0 = const()[name = tensor<string, []>("op_5410_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5410_pad_0 = const()[name = tensor<string, []>("op_5410_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5410_dilations_0 = const()[name = tensor<string, []>("op_5410_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5410_groups_0 = const()[name = tensor<string, []>("op_5410_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_22_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142053824))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142578176))), name = tensor<string, []>("blocks_22_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_22_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142578304)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5410_cast_fp16 = conv(bias = blocks_22_attn_value_bias_to_fp16, dilations = var_5410_dilations_0, groups = var_5410_groups_0, pad = var_5410_pad_0, pad_type = var_5410_pad_type_0, strides = var_5410_strides_0, weight = blocks_22_attn_value_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor<string, []>("op_5410_cast_fp16")];
            tensor<int32, [16]> tile_66 = const()[name = tensor<string, []>("tile_66"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5413_axis_0 = const()[name = tensor<string, []>("op_5413_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5413_cast_fp16_15 = split(axis = var_5413_axis_0, split_sizes = tile_66, x = var_5412_cast_fp16)[name = tensor<string, []>("op_5413_cast_fp16")];
            tensor<int32, [4]> var_5430_perm_0 = const()[name = tensor<string, []>("op_5430_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_67 = const()[name = tensor<string, []>("tile_67"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5431_axis_0 = const()[name = tensor<string, []>("op_5431_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_5430_cast_fp16 = transpose(perm = var_5430_perm_0, x = k_45_cast_fp16)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_5431_cast_fp16_15 = split(axis = var_5431_axis_0, split_sizes = tile_67, x = var_5430_cast_fp16)[name = tensor<string, []>("op_5431_cast_fp16")];
            tensor<int32, [16]> tile_68 = const()[name = tensor<string, []>("tile_68"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5448_axis_0 = const()[name = tensor<string, []>("op_5448_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5448_cast_fp16_15 = split(axis = var_5448_axis_0, split_sizes = tile_68, x = var_5410_cast_fp16)[name = tensor<string, []>("op_5448_cast_fp16")];
            tensor<string, []> aw_705_equation_0 = const()[name = tensor<string, []>("aw_705_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_705_cast_fp16 = einsum(equation = aw_705_equation_0, values = (var_5431_cast_fp16_0, var_5413_cast_fp16_0))[name = tensor<string, []>("aw_705_cast_fp16")];
            tensor<string, []> aw_707_equation_0 = const()[name = tensor<string, []>("aw_707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_707_cast_fp16 = einsum(equation = aw_707_equation_0, values = (var_5431_cast_fp16_1, var_5413_cast_fp16_1))[name = tensor<string, []>("aw_707_cast_fp16")];
            tensor<string, []> aw_709_equation_0 = const()[name = tensor<string, []>("aw_709_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_709_cast_fp16 = einsum(equation = aw_709_equation_0, values = (var_5431_cast_fp16_2, var_5413_cast_fp16_2))[name = tensor<string, []>("aw_709_cast_fp16")];
            tensor<string, []> aw_711_equation_0 = const()[name = tensor<string, []>("aw_711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_711_cast_fp16 = einsum(equation = aw_711_equation_0, values = (var_5431_cast_fp16_3, var_5413_cast_fp16_3))[name = tensor<string, []>("aw_711_cast_fp16")];
            tensor<string, []> aw_713_equation_0 = const()[name = tensor<string, []>("aw_713_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_713_cast_fp16 = einsum(equation = aw_713_equation_0, values = (var_5431_cast_fp16_4, var_5413_cast_fp16_4))[name = tensor<string, []>("aw_713_cast_fp16")];
            tensor<string, []> aw_715_equation_0 = const()[name = tensor<string, []>("aw_715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_715_cast_fp16 = einsum(equation = aw_715_equation_0, values = (var_5431_cast_fp16_5, var_5413_cast_fp16_5))[name = tensor<string, []>("aw_715_cast_fp16")];
            tensor<string, []> aw_717_equation_0 = const()[name = tensor<string, []>("aw_717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_717_cast_fp16 = einsum(equation = aw_717_equation_0, values = (var_5431_cast_fp16_6, var_5413_cast_fp16_6))[name = tensor<string, []>("aw_717_cast_fp16")];
            tensor<string, []> aw_719_equation_0 = const()[name = tensor<string, []>("aw_719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_719_cast_fp16 = einsum(equation = aw_719_equation_0, values = (var_5431_cast_fp16_7, var_5413_cast_fp16_7))[name = tensor<string, []>("aw_719_cast_fp16")];
            tensor<string, []> aw_721_equation_0 = const()[name = tensor<string, []>("aw_721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_721_cast_fp16 = einsum(equation = aw_721_equation_0, values = (var_5431_cast_fp16_8, var_5413_cast_fp16_8))[name = tensor<string, []>("aw_721_cast_fp16")];
            tensor<string, []> aw_723_equation_0 = const()[name = tensor<string, []>("aw_723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_723_cast_fp16 = einsum(equation = aw_723_equation_0, values = (var_5431_cast_fp16_9, var_5413_cast_fp16_9))[name = tensor<string, []>("aw_723_cast_fp16")];
            tensor<string, []> aw_725_equation_0 = const()[name = tensor<string, []>("aw_725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_725_cast_fp16 = einsum(equation = aw_725_equation_0, values = (var_5431_cast_fp16_10, var_5413_cast_fp16_10))[name = tensor<string, []>("aw_725_cast_fp16")];
            tensor<string, []> aw_727_equation_0 = const()[name = tensor<string, []>("aw_727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_727_cast_fp16 = einsum(equation = aw_727_equation_0, values = (var_5431_cast_fp16_11, var_5413_cast_fp16_11))[name = tensor<string, []>("aw_727_cast_fp16")];
            tensor<string, []> aw_729_equation_0 = const()[name = tensor<string, []>("aw_729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_729_cast_fp16 = einsum(equation = aw_729_equation_0, values = (var_5431_cast_fp16_12, var_5413_cast_fp16_12))[name = tensor<string, []>("aw_729_cast_fp16")];
            tensor<string, []> aw_731_equation_0 = const()[name = tensor<string, []>("aw_731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_731_cast_fp16 = einsum(equation = aw_731_equation_0, values = (var_5431_cast_fp16_13, var_5413_cast_fp16_13))[name = tensor<string, []>("aw_731_cast_fp16")];
            tensor<string, []> aw_733_equation_0 = const()[name = tensor<string, []>("aw_733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_733_cast_fp16 = einsum(equation = aw_733_equation_0, values = (var_5431_cast_fp16_14, var_5413_cast_fp16_14))[name = tensor<string, []>("aw_733_cast_fp16")];
            tensor<string, []> aw_735_equation_0 = const()[name = tensor<string, []>("aw_735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_735_cast_fp16 = einsum(equation = aw_735_equation_0, values = (var_5431_cast_fp16_15, var_5413_cast_fp16_15))[name = tensor<string, []>("aw_735_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5497_cast_fp16 = softmax(axis = var_5361, x = aw_705_cast_fp16)[name = tensor<string, []>("op_5497_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5498_cast_fp16 = softmax(axis = var_5361, x = aw_707_cast_fp16)[name = tensor<string, []>("op_5498_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5499_cast_fp16 = softmax(axis = var_5361, x = aw_709_cast_fp16)[name = tensor<string, []>("op_5499_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5500_cast_fp16 = softmax(axis = var_5361, x = aw_711_cast_fp16)[name = tensor<string, []>("op_5500_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5501_cast_fp16 = softmax(axis = var_5361, x = aw_713_cast_fp16)[name = tensor<string, []>("op_5501_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5502_cast_fp16 = softmax(axis = var_5361, x = aw_715_cast_fp16)[name = tensor<string, []>("op_5502_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5503_cast_fp16 = softmax(axis = var_5361, x = aw_717_cast_fp16)[name = tensor<string, []>("op_5503_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5504_cast_fp16 = softmax(axis = var_5361, x = aw_719_cast_fp16)[name = tensor<string, []>("op_5504_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5505_cast_fp16 = softmax(axis = var_5361, x = aw_721_cast_fp16)[name = tensor<string, []>("op_5505_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5506_cast_fp16 = softmax(axis = var_5361, x = aw_723_cast_fp16)[name = tensor<string, []>("op_5506_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5507_cast_fp16 = softmax(axis = var_5361, x = aw_725_cast_fp16)[name = tensor<string, []>("op_5507_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5508_cast_fp16 = softmax(axis = var_5361, x = aw_727_cast_fp16)[name = tensor<string, []>("op_5508_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5509_cast_fp16 = softmax(axis = var_5361, x = aw_729_cast_fp16)[name = tensor<string, []>("op_5509_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5510_cast_fp16 = softmax(axis = var_5361, x = aw_731_cast_fp16)[name = tensor<string, []>("op_5510_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5511_cast_fp16 = softmax(axis = var_5361, x = aw_733_cast_fp16)[name = tensor<string, []>("op_5511_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5512_cast_fp16 = softmax(axis = var_5361, x = aw_735_cast_fp16)[name = tensor<string, []>("op_5512_cast_fp16")];
            tensor<string, []> var_5514_equation_0 = const()[name = tensor<string, []>("op_5514_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5514_cast_fp16 = einsum(equation = var_5514_equation_0, values = (var_5448_cast_fp16_0, var_5497_cast_fp16))[name = tensor<string, []>("op_5514_cast_fp16")];
            tensor<string, []> var_5516_equation_0 = const()[name = tensor<string, []>("op_5516_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5516_cast_fp16 = einsum(equation = var_5516_equation_0, values = (var_5448_cast_fp16_1, var_5498_cast_fp16))[name = tensor<string, []>("op_5516_cast_fp16")];
            tensor<string, []> var_5518_equation_0 = const()[name = tensor<string, []>("op_5518_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5518_cast_fp16 = einsum(equation = var_5518_equation_0, values = (var_5448_cast_fp16_2, var_5499_cast_fp16))[name = tensor<string, []>("op_5518_cast_fp16")];
            tensor<string, []> var_5520_equation_0 = const()[name = tensor<string, []>("op_5520_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5520_cast_fp16 = einsum(equation = var_5520_equation_0, values = (var_5448_cast_fp16_3, var_5500_cast_fp16))[name = tensor<string, []>("op_5520_cast_fp16")];
            tensor<string, []> var_5522_equation_0 = const()[name = tensor<string, []>("op_5522_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5522_cast_fp16 = einsum(equation = var_5522_equation_0, values = (var_5448_cast_fp16_4, var_5501_cast_fp16))[name = tensor<string, []>("op_5522_cast_fp16")];
            tensor<string, []> var_5524_equation_0 = const()[name = tensor<string, []>("op_5524_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5524_cast_fp16 = einsum(equation = var_5524_equation_0, values = (var_5448_cast_fp16_5, var_5502_cast_fp16))[name = tensor<string, []>("op_5524_cast_fp16")];
            tensor<string, []> var_5526_equation_0 = const()[name = tensor<string, []>("op_5526_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5526_cast_fp16 = einsum(equation = var_5526_equation_0, values = (var_5448_cast_fp16_6, var_5503_cast_fp16))[name = tensor<string, []>("op_5526_cast_fp16")];
            tensor<string, []> var_5528_equation_0 = const()[name = tensor<string, []>("op_5528_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5528_cast_fp16 = einsum(equation = var_5528_equation_0, values = (var_5448_cast_fp16_7, var_5504_cast_fp16))[name = tensor<string, []>("op_5528_cast_fp16")];
            tensor<string, []> var_5530_equation_0 = const()[name = tensor<string, []>("op_5530_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5530_cast_fp16 = einsum(equation = var_5530_equation_0, values = (var_5448_cast_fp16_8, var_5505_cast_fp16))[name = tensor<string, []>("op_5530_cast_fp16")];
            tensor<string, []> var_5532_equation_0 = const()[name = tensor<string, []>("op_5532_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5532_cast_fp16 = einsum(equation = var_5532_equation_0, values = (var_5448_cast_fp16_9, var_5506_cast_fp16))[name = tensor<string, []>("op_5532_cast_fp16")];
            tensor<string, []> var_5534_equation_0 = const()[name = tensor<string, []>("op_5534_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5534_cast_fp16 = einsum(equation = var_5534_equation_0, values = (var_5448_cast_fp16_10, var_5507_cast_fp16))[name = tensor<string, []>("op_5534_cast_fp16")];
            tensor<string, []> var_5536_equation_0 = const()[name = tensor<string, []>("op_5536_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5536_cast_fp16 = einsum(equation = var_5536_equation_0, values = (var_5448_cast_fp16_11, var_5508_cast_fp16))[name = tensor<string, []>("op_5536_cast_fp16")];
            tensor<string, []> var_5538_equation_0 = const()[name = tensor<string, []>("op_5538_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5538_cast_fp16 = einsum(equation = var_5538_equation_0, values = (var_5448_cast_fp16_12, var_5509_cast_fp16))[name = tensor<string, []>("op_5538_cast_fp16")];
            tensor<string, []> var_5540_equation_0 = const()[name = tensor<string, []>("op_5540_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5540_cast_fp16 = einsum(equation = var_5540_equation_0, values = (var_5448_cast_fp16_13, var_5510_cast_fp16))[name = tensor<string, []>("op_5540_cast_fp16")];
            tensor<string, []> var_5542_equation_0 = const()[name = tensor<string, []>("op_5542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5542_cast_fp16 = einsum(equation = var_5542_equation_0, values = (var_5448_cast_fp16_14, var_5511_cast_fp16))[name = tensor<string, []>("op_5542_cast_fp16")];
            tensor<string, []> var_5544_equation_0 = const()[name = tensor<string, []>("op_5544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5544_cast_fp16 = einsum(equation = var_5544_equation_0, values = (var_5448_cast_fp16_15, var_5512_cast_fp16))[name = tensor<string, []>("op_5544_cast_fp16")];
            tensor<bool, []> input_227_interleave_0 = const()[name = tensor<string, []>("input_227_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_227_cast_fp16 = concat(axis = var_5361, interleave = input_227_interleave_0, values = (var_5514_cast_fp16, var_5516_cast_fp16, var_5518_cast_fp16, var_5520_cast_fp16, var_5522_cast_fp16, var_5524_cast_fp16, var_5526_cast_fp16, var_5528_cast_fp16, var_5530_cast_fp16, var_5532_cast_fp16, var_5534_cast_fp16, var_5536_cast_fp16, var_5538_cast_fp16, var_5540_cast_fp16, var_5542_cast_fp16, var_5544_cast_fp16))[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<string, []> var_5553_pad_type_0 = const()[name = tensor<string, []>("op_5553_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5553_strides_0 = const()[name = tensor<string, []>("op_5553_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5553_pad_0 = const()[name = tensor<string, []>("op_5553_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5553_dilations_0 = const()[name = tensor<string, []>("op_5553_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5553_groups_0 = const()[name = tensor<string, []>("op_5553_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_22_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142580416))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143104768))), name = tensor<string, []>("blocks_22_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_22_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143104896)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5553_cast_fp16 = conv(bias = blocks_22_attn_out_bias_to_fp16, dilations = var_5553_dilations_0, groups = var_5553_groups_0, pad = var_5553_pad_0, pad_type = var_5553_pad_type_0, strides = var_5553_strides_0, weight = blocks_22_attn_out_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = tensor<string, []>("op_5553_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = var_5553_cast_fp16)[name = tensor<string, []>("inputs_91_cast_fp16")];
            tensor<int32, [1]> input_229_axes_0 = const()[name = tensor<string, []>("input_229_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_229_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_229_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143107008)))];
            tensor<fp16, [1024]> input_229_beta_0_to_fp16 = const()[name = tensor<string, []>("input_229_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143109120)))];
            tensor<fp16, []> var_5563_to_fp16 = const()[name = tensor<string, []>("op_5563_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = input_229_beta_0_to_fp16, epsilon = var_5563_to_fp16, gamma = input_229_gamma_0_to_fp16, x = inputs_91_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<string, []> input_231_pad_type_0 = const()[name = tensor<string, []>("input_231_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_231_strides_0 = const()[name = tensor<string, []>("input_231_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_231_pad_0 = const()[name = tensor<string, []>("input_231_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_231_dilations_0 = const()[name = tensor<string, []>("input_231_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_231_groups_0 = const()[name = tensor<string, []>("input_231_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_22_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143111232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145208448))), name = tensor<string, []>("blocks_22_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_22_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145208576)))];
            tensor<fp16, [1, 4096, 1, 600]> input_231_cast_fp16 = conv(bias = blocks_22_mlp_0_bias_to_fp16, dilations = input_231_dilations_0, groups = input_231_groups_0, pad = input_231_pad_0, pad_type = input_231_pad_type_0, strides = input_231_strides_0, weight = blocks_22_mlp_0_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")];
            tensor<string, []> input_233_mode_0 = const()[name = tensor<string, []>("input_233_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_233_cast_fp16 = gelu(mode = input_233_mode_0, x = input_231_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")];
            tensor<string, []> var_5589_pad_type_0 = const()[name = tensor<string, []>("op_5589_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5589_strides_0 = const()[name = tensor<string, []>("op_5589_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5589_pad_0 = const()[name = tensor<string, []>("op_5589_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5589_dilations_0 = const()[name = tensor<string, []>("op_5589_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5589_groups_0 = const()[name = tensor<string, []>("op_5589_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_22_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145216832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147314048))), name = tensor<string, []>("blocks_22_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_22_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147314176)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5589_cast_fp16 = conv(bias = blocks_22_mlp_2_bias_to_fp16, dilations = var_5589_dilations_0, groups = var_5589_groups_0, pad = var_5589_pad_0, pad_type = var_5589_pad_type_0, strides = var_5589_strides_0, weight = blocks_22_mlp_2_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor<string, []>("op_5589_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = var_5589_cast_fp16)[name = tensor<string, []>("inputs_93_cast_fp16")];
            tensor<int32, []> var_5598 = const()[name = tensor<string, []>("op_5598"), val = tensor<int32, []>(1)];
            tensor<int32, [1]> input_235_axes_0 = const()[name = tensor<string, []>("input_235_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_235_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_235_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147316288)))];
            tensor<fp16, [1024]> input_235_beta_0_to_fp16 = const()[name = tensor<string, []>("input_235_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147318400)))];
            tensor<fp16, []> var_5614_to_fp16 = const()[name = tensor<string, []>("op_5614_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_235_cast_fp16 = layer_norm(axes = input_235_axes_0, beta = input_235_beta_0_to_fp16, epsilon = var_5614_to_fp16, gamma = input_235_gamma_0_to_fp16, x = inputs_93_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")];
            tensor<string, []> q_47_pad_type_0 = const()[name = tensor<string, []>("q_47_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> q_47_strides_0 = const()[name = tensor<string, []>("q_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_47_pad_0 = const()[name = tensor<string, []>("q_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_47_dilations_0 = const()[name = tensor<string, []>("q_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> q_47_groups_0 = const()[name = tensor<string, []>("q_47_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> op_5649_weight_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147320512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147844864))), name = tensor<string, []>("op_5649_weight_0_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> var_5649_bias_0_to_fp16 = const()[name = tensor<string, []>("op_5649_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147844992)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5649_cast_fp16 = conv(bias = var_5649_bias_0_to_fp16, dilations = q_47_dilations_0, groups = q_47_groups_0, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = q_47_strides_0, weight = op_5649_weight_0_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor<string, []>("op_5649_cast_fp16")];
            tensor<string, []> k_47_pad_type_0 = const()[name = tensor<string, []>("k_47_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> k_47_strides_0 = const()[name = tensor<string, []>("k_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_47_pad_0 = const()[name = tensor<string, []>("k_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_47_dilations_0 = const()[name = tensor<string, []>("k_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> k_47_groups_0 = const()[name = tensor<string, []>("k_47_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_23_attn_key_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147847104))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148371456))), name = tensor<string, []>("blocks_23_attn_key_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1, 1024, 1, 600]> k_47_cast_fp16 = conv(dilations = k_47_dilations_0, groups = k_47_groups_0, pad = k_47_pad_0, pad_type = k_47_pad_type_0, strides = k_47_strides_0, weight = blocks_23_attn_key_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor<string, []>("k_47_cast_fp16")];
            tensor<string, []> var_5647_pad_type_0 = const()[name = tensor<string, []>("op_5647_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5647_strides_0 = const()[name = tensor<string, []>("op_5647_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5647_pad_0 = const()[name = tensor<string, []>("op_5647_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5647_dilations_0 = const()[name = tensor<string, []>("op_5647_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5647_groups_0 = const()[name = tensor<string, []>("op_5647_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_23_attn_value_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148371584))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148895936))), name = tensor<string, []>("blocks_23_attn_value_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_23_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_value_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148896064)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5647_cast_fp16 = conv(bias = blocks_23_attn_value_bias_to_fp16, dilations = var_5647_dilations_0, groups = var_5647_groups_0, pad = var_5647_pad_0, pad_type = var_5647_pad_type_0, strides = var_5647_strides_0, weight = blocks_23_attn_value_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = tensor<string, []>("op_5647_cast_fp16")];
            tensor<int32, [16]> tile_69 = const()[name = tensor<string, []>("tile_69"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5650_axis_0 = const()[name = tensor<string, []>("op_5650_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5650_cast_fp16_15 = split(axis = var_5650_axis_0, split_sizes = tile_69, x = var_5649_cast_fp16)[name = tensor<string, []>("op_5650_cast_fp16")];
            tensor<int32, [4]> var_5667_perm_0 = const()[name = tensor<string, []>("op_5667_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [16]> tile_70 = const()[name = tensor<string, []>("tile_70"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5668_axis_0 = const()[name = tensor<string, []>("op_5668_axis_0"), val = tensor<int32, []>(3)];
            tensor<fp16, [1, 600, 1, 1024]> var_5667_cast_fp16 = transpose(perm = var_5667_perm_0, x = k_47_cast_fp16)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_0, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_1, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_2, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_3, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_4, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_5, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_6, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_7, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_8, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_9, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_10, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_11, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_12, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_13, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_14, tensor<fp16, [1, 600, 1, 64]> var_5668_cast_fp16_15 = split(axis = var_5668_axis_0, split_sizes = tile_70, x = var_5667_cast_fp16)[name = tensor<string, []>("op_5668_cast_fp16")];
            tensor<int32, [16]> tile_71 = const()[name = tensor<string, []>("tile_71"), val = tensor<int32, [16]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])];
            tensor<int32, []> var_5685_axis_0 = const()[name = tensor<string, []>("op_5685_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_0, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_1, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_2, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_3, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_4, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_5, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_6, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_7, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_8, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_9, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_10, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_11, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_12, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_13, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_14, tensor<fp16, [1, 64, 1, 600]> var_5685_cast_fp16_15 = split(axis = var_5685_axis_0, split_sizes = tile_71, x = var_5647_cast_fp16)[name = tensor<string, []>("op_5685_cast_fp16")];
            tensor<string, []> aw_737_equation_0 = const()[name = tensor<string, []>("aw_737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_737_cast_fp16 = einsum(equation = aw_737_equation_0, values = (var_5668_cast_fp16_0, var_5650_cast_fp16_0))[name = tensor<string, []>("aw_737_cast_fp16")];
            tensor<string, []> aw_739_equation_0 = const()[name = tensor<string, []>("aw_739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_739_cast_fp16 = einsum(equation = aw_739_equation_0, values = (var_5668_cast_fp16_1, var_5650_cast_fp16_1))[name = tensor<string, []>("aw_739_cast_fp16")];
            tensor<string, []> aw_741_equation_0 = const()[name = tensor<string, []>("aw_741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_741_cast_fp16 = einsum(equation = aw_741_equation_0, values = (var_5668_cast_fp16_2, var_5650_cast_fp16_2))[name = tensor<string, []>("aw_741_cast_fp16")];
            tensor<string, []> aw_743_equation_0 = const()[name = tensor<string, []>("aw_743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_743_cast_fp16 = einsum(equation = aw_743_equation_0, values = (var_5668_cast_fp16_3, var_5650_cast_fp16_3))[name = tensor<string, []>("aw_743_cast_fp16")];
            tensor<string, []> aw_745_equation_0 = const()[name = tensor<string, []>("aw_745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_745_cast_fp16 = einsum(equation = aw_745_equation_0, values = (var_5668_cast_fp16_4, var_5650_cast_fp16_4))[name = tensor<string, []>("aw_745_cast_fp16")];
            tensor<string, []> aw_747_equation_0 = const()[name = tensor<string, []>("aw_747_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_747_cast_fp16 = einsum(equation = aw_747_equation_0, values = (var_5668_cast_fp16_5, var_5650_cast_fp16_5))[name = tensor<string, []>("aw_747_cast_fp16")];
            tensor<string, []> aw_749_equation_0 = const()[name = tensor<string, []>("aw_749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_749_cast_fp16 = einsum(equation = aw_749_equation_0, values = (var_5668_cast_fp16_6, var_5650_cast_fp16_6))[name = tensor<string, []>("aw_749_cast_fp16")];
            tensor<string, []> aw_751_equation_0 = const()[name = tensor<string, []>("aw_751_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_751_cast_fp16 = einsum(equation = aw_751_equation_0, values = (var_5668_cast_fp16_7, var_5650_cast_fp16_7))[name = tensor<string, []>("aw_751_cast_fp16")];
            tensor<string, []> aw_753_equation_0 = const()[name = tensor<string, []>("aw_753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_753_cast_fp16 = einsum(equation = aw_753_equation_0, values = (var_5668_cast_fp16_8, var_5650_cast_fp16_8))[name = tensor<string, []>("aw_753_cast_fp16")];
            tensor<string, []> aw_755_equation_0 = const()[name = tensor<string, []>("aw_755_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_755_cast_fp16 = einsum(equation = aw_755_equation_0, values = (var_5668_cast_fp16_9, var_5650_cast_fp16_9))[name = tensor<string, []>("aw_755_cast_fp16")];
            tensor<string, []> aw_757_equation_0 = const()[name = tensor<string, []>("aw_757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_757_cast_fp16 = einsum(equation = aw_757_equation_0, values = (var_5668_cast_fp16_10, var_5650_cast_fp16_10))[name = tensor<string, []>("aw_757_cast_fp16")];
            tensor<string, []> aw_759_equation_0 = const()[name = tensor<string, []>("aw_759_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_759_cast_fp16 = einsum(equation = aw_759_equation_0, values = (var_5668_cast_fp16_11, var_5650_cast_fp16_11))[name = tensor<string, []>("aw_759_cast_fp16")];
            tensor<string, []> aw_761_equation_0 = const()[name = tensor<string, []>("aw_761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_761_cast_fp16 = einsum(equation = aw_761_equation_0, values = (var_5668_cast_fp16_12, var_5650_cast_fp16_12))[name = tensor<string, []>("aw_761_cast_fp16")];
            tensor<string, []> aw_763_equation_0 = const()[name = tensor<string, []>("aw_763_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_763_cast_fp16 = einsum(equation = aw_763_equation_0, values = (var_5668_cast_fp16_13, var_5650_cast_fp16_13))[name = tensor<string, []>("aw_763_cast_fp16")];
            tensor<string, []> aw_765_equation_0 = const()[name = tensor<string, []>("aw_765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_765_cast_fp16 = einsum(equation = aw_765_equation_0, values = (var_5668_cast_fp16_14, var_5650_cast_fp16_14))[name = tensor<string, []>("aw_765_cast_fp16")];
            tensor<string, []> aw_equation_0 = const()[name = tensor<string, []>("aw_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 600, 1, 600]> aw_cast_fp16 = einsum(equation = aw_equation_0, values = (var_5668_cast_fp16_15, var_5650_cast_fp16_15))[name = tensor<string, []>("aw_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5734_cast_fp16 = softmax(axis = var_5598, x = aw_737_cast_fp16)[name = tensor<string, []>("op_5734_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5735_cast_fp16 = softmax(axis = var_5598, x = aw_739_cast_fp16)[name = tensor<string, []>("op_5735_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5736_cast_fp16 = softmax(axis = var_5598, x = aw_741_cast_fp16)[name = tensor<string, []>("op_5736_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5737_cast_fp16 = softmax(axis = var_5598, x = aw_743_cast_fp16)[name = tensor<string, []>("op_5737_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5738_cast_fp16 = softmax(axis = var_5598, x = aw_745_cast_fp16)[name = tensor<string, []>("op_5738_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5739_cast_fp16 = softmax(axis = var_5598, x = aw_747_cast_fp16)[name = tensor<string, []>("op_5739_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5740_cast_fp16 = softmax(axis = var_5598, x = aw_749_cast_fp16)[name = tensor<string, []>("op_5740_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5741_cast_fp16 = softmax(axis = var_5598, x = aw_751_cast_fp16)[name = tensor<string, []>("op_5741_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5742_cast_fp16 = softmax(axis = var_5598, x = aw_753_cast_fp16)[name = tensor<string, []>("op_5742_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5743_cast_fp16 = softmax(axis = var_5598, x = aw_755_cast_fp16)[name = tensor<string, []>("op_5743_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5744_cast_fp16 = softmax(axis = var_5598, x = aw_757_cast_fp16)[name = tensor<string, []>("op_5744_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5745_cast_fp16 = softmax(axis = var_5598, x = aw_759_cast_fp16)[name = tensor<string, []>("op_5745_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5746_cast_fp16 = softmax(axis = var_5598, x = aw_761_cast_fp16)[name = tensor<string, []>("op_5746_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5747_cast_fp16 = softmax(axis = var_5598, x = aw_763_cast_fp16)[name = tensor<string, []>("op_5747_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5748_cast_fp16 = softmax(axis = var_5598, x = aw_765_cast_fp16)[name = tensor<string, []>("op_5748_cast_fp16")];
            tensor<fp16, [1, 600, 1, 600]> var_5749_cast_fp16 = softmax(axis = var_5598, x = aw_cast_fp16)[name = tensor<string, []>("op_5749_cast_fp16")];
            tensor<string, []> var_5751_equation_0 = const()[name = tensor<string, []>("op_5751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5751_cast_fp16 = einsum(equation = var_5751_equation_0, values = (var_5685_cast_fp16_0, var_5734_cast_fp16))[name = tensor<string, []>("op_5751_cast_fp16")];
            tensor<string, []> var_5753_equation_0 = const()[name = tensor<string, []>("op_5753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5753_cast_fp16 = einsum(equation = var_5753_equation_0, values = (var_5685_cast_fp16_1, var_5735_cast_fp16))[name = tensor<string, []>("op_5753_cast_fp16")];
            tensor<string, []> var_5755_equation_0 = const()[name = tensor<string, []>("op_5755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5755_cast_fp16 = einsum(equation = var_5755_equation_0, values = (var_5685_cast_fp16_2, var_5736_cast_fp16))[name = tensor<string, []>("op_5755_cast_fp16")];
            tensor<string, []> var_5757_equation_0 = const()[name = tensor<string, []>("op_5757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5757_cast_fp16 = einsum(equation = var_5757_equation_0, values = (var_5685_cast_fp16_3, var_5737_cast_fp16))[name = tensor<string, []>("op_5757_cast_fp16")];
            tensor<string, []> var_5759_equation_0 = const()[name = tensor<string, []>("op_5759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5759_cast_fp16 = einsum(equation = var_5759_equation_0, values = (var_5685_cast_fp16_4, var_5738_cast_fp16))[name = tensor<string, []>("op_5759_cast_fp16")];
            tensor<string, []> var_5761_equation_0 = const()[name = tensor<string, []>("op_5761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5761_cast_fp16 = einsum(equation = var_5761_equation_0, values = (var_5685_cast_fp16_5, var_5739_cast_fp16))[name = tensor<string, []>("op_5761_cast_fp16")];
            tensor<string, []> var_5763_equation_0 = const()[name = tensor<string, []>("op_5763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5763_cast_fp16 = einsum(equation = var_5763_equation_0, values = (var_5685_cast_fp16_6, var_5740_cast_fp16))[name = tensor<string, []>("op_5763_cast_fp16")];
            tensor<string, []> var_5765_equation_0 = const()[name = tensor<string, []>("op_5765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5765_cast_fp16 = einsum(equation = var_5765_equation_0, values = (var_5685_cast_fp16_7, var_5741_cast_fp16))[name = tensor<string, []>("op_5765_cast_fp16")];
            tensor<string, []> var_5767_equation_0 = const()[name = tensor<string, []>("op_5767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5767_cast_fp16 = einsum(equation = var_5767_equation_0, values = (var_5685_cast_fp16_8, var_5742_cast_fp16))[name = tensor<string, []>("op_5767_cast_fp16")];
            tensor<string, []> var_5769_equation_0 = const()[name = tensor<string, []>("op_5769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5769_cast_fp16 = einsum(equation = var_5769_equation_0, values = (var_5685_cast_fp16_9, var_5743_cast_fp16))[name = tensor<string, []>("op_5769_cast_fp16")];
            tensor<string, []> var_5771_equation_0 = const()[name = tensor<string, []>("op_5771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5771_cast_fp16 = einsum(equation = var_5771_equation_0, values = (var_5685_cast_fp16_10, var_5744_cast_fp16))[name = tensor<string, []>("op_5771_cast_fp16")];
            tensor<string, []> var_5773_equation_0 = const()[name = tensor<string, []>("op_5773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5773_cast_fp16 = einsum(equation = var_5773_equation_0, values = (var_5685_cast_fp16_11, var_5745_cast_fp16))[name = tensor<string, []>("op_5773_cast_fp16")];
            tensor<string, []> var_5775_equation_0 = const()[name = tensor<string, []>("op_5775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5775_cast_fp16 = einsum(equation = var_5775_equation_0, values = (var_5685_cast_fp16_12, var_5746_cast_fp16))[name = tensor<string, []>("op_5775_cast_fp16")];
            tensor<string, []> var_5777_equation_0 = const()[name = tensor<string, []>("op_5777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5777_cast_fp16 = einsum(equation = var_5777_equation_0, values = (var_5685_cast_fp16_13, var_5747_cast_fp16))[name = tensor<string, []>("op_5777_cast_fp16")];
            tensor<string, []> var_5779_equation_0 = const()[name = tensor<string, []>("op_5779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5779_cast_fp16 = einsum(equation = var_5779_equation_0, values = (var_5685_cast_fp16_14, var_5748_cast_fp16))[name = tensor<string, []>("op_5779_cast_fp16")];
            tensor<string, []> var_5781_equation_0 = const()[name = tensor<string, []>("op_5781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 600]> var_5781_cast_fp16 = einsum(equation = var_5781_equation_0, values = (var_5685_cast_fp16_15, var_5749_cast_fp16))[name = tensor<string, []>("op_5781_cast_fp16")];
            tensor<bool, []> input_237_interleave_0 = const()[name = tensor<string, []>("input_237_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 1024, 1, 600]> input_237_cast_fp16 = concat(axis = var_5598, interleave = input_237_interleave_0, values = (var_5751_cast_fp16, var_5753_cast_fp16, var_5755_cast_fp16, var_5757_cast_fp16, var_5759_cast_fp16, var_5761_cast_fp16, var_5763_cast_fp16, var_5765_cast_fp16, var_5767_cast_fp16, var_5769_cast_fp16, var_5771_cast_fp16, var_5773_cast_fp16, var_5775_cast_fp16, var_5777_cast_fp16, var_5779_cast_fp16, var_5781_cast_fp16))[name = tensor<string, []>("input_237_cast_fp16")];
            tensor<string, []> var_5790_pad_type_0 = const()[name = tensor<string, []>("op_5790_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5790_strides_0 = const()[name = tensor<string, []>("op_5790_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5790_pad_0 = const()[name = tensor<string, []>("op_5790_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5790_dilations_0 = const()[name = tensor<string, []>("op_5790_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5790_groups_0 = const()[name = tensor<string, []>("op_5790_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 1024, 1, 1]> blocks_23_attn_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148898176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149422528))), name = tensor<string, []>("blocks_23_attn_out_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 1024, 1, 1])];
            tensor<fp16, [1024]> blocks_23_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149422656)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5790_cast_fp16 = conv(bias = blocks_23_attn_out_bias_to_fp16, dilations = var_5790_dilations_0, groups = var_5790_groups_0, pad = var_5790_pad_0, pad_type = var_5790_pad_type_0, strides = var_5790_strides_0, weight = blocks_23_attn_out_weight_to_fp16_palettized, x = input_237_cast_fp16)[name = tensor<string, []>("op_5790_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = var_5790_cast_fp16)[name = tensor<string, []>("inputs_95_cast_fp16")];
            tensor<int32, [1]> input_239_axes_0 = const()[name = tensor<string, []>("input_239_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> input_239_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_239_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149424768)))];
            tensor<fp16, [1024]> input_239_beta_0_to_fp16 = const()[name = tensor<string, []>("input_239_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149426880)))];
            tensor<fp16, []> var_5800_to_fp16 = const()[name = tensor<string, []>("op_5800_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> input_239_cast_fp16 = layer_norm(axes = input_239_axes_0, beta = input_239_beta_0_to_fp16, epsilon = var_5800_to_fp16, gamma = input_239_gamma_0_to_fp16, x = inputs_95_cast_fp16)[name = tensor<string, []>("input_239_cast_fp16")];
            tensor<string, []> input_241_pad_type_0 = const()[name = tensor<string, []>("input_241_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> input_241_strides_0 = const()[name = tensor<string, []>("input_241_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_241_pad_0 = const()[name = tensor<string, []>("input_241_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_241_dilations_0 = const()[name = tensor<string, []>("input_241_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> input_241_groups_0 = const()[name = tensor<string, []>("input_241_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [4096, 1024, 1, 1]> blocks_23_mlp_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149428992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151526208))), name = tensor<string, []>("blocks_23_mlp_0_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([4096, 1024, 1, 1])];
            tensor<fp16, [4096]> blocks_23_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_0_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151526336)))];
            tensor<fp16, [1, 4096, 1, 600]> input_241_cast_fp16 = conv(bias = blocks_23_mlp_0_bias_to_fp16, dilations = input_241_dilations_0, groups = input_241_groups_0, pad = input_241_pad_0, pad_type = input_241_pad_type_0, strides = input_241_strides_0, weight = blocks_23_mlp_0_weight_to_fp16_palettized, x = input_239_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<string, []> input_243_mode_0 = const()[name = tensor<string, []>("input_243_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 4096, 1, 600]> input_243_cast_fp16 = gelu(mode = input_243_mode_0, x = input_241_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")];
            tensor<string, []> var_5826_pad_type_0 = const()[name = tensor<string, []>("op_5826_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> var_5826_strides_0 = const()[name = tensor<string, []>("op_5826_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5826_pad_0 = const()[name = tensor<string, []>("op_5826_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5826_dilations_0 = const()[name = tensor<string, []>("op_5826_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> var_5826_groups_0 = const()[name = tensor<string, []>("op_5826_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 4096, 1, 1]> blocks_23_mlp_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151534592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153631808))), name = tensor<string, []>("blocks_23_mlp_2_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1024, 4096, 1, 1])];
            tensor<fp16, [1024]> blocks_23_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153631936)))];
            tensor<fp16, [1, 1024, 1, 600]> var_5826_cast_fp16 = conv(bias = blocks_23_mlp_2_bias_to_fp16, dilations = var_5826_dilations_0, groups = var_5826_groups_0, pad = var_5826_pad_0, pad_type = var_5826_pad_type_0, strides = var_5826_strides_0, weight = blocks_23_mlp_2_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = tensor<string, []>("op_5826_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 600]> inputs_cast_fp16 = add(x = inputs_95_cast_fp16, y = var_5826_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")];
            tensor<int32, [1]> x_5_axes_0 = const()[name = tensor<string, []>("x_5_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024]> x_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("x_5_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153634048)))];
            tensor<fp16, [1024]> x_5_beta_0_to_fp16 = const()[name = tensor<string, []>("x_5_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153636160)))];
            tensor<fp16, []> var_5840_to_fp16 = const()[name = tensor<string, []>("op_5840_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 600]> x_5_cast_fp16 = layer_norm(axes = x_5_axes_0, beta = x_5_beta_0_to_fp16, epsilon = var_5840_to_fp16, gamma = x_5_gamma_0_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("x_5_cast_fp16")];
            tensor<int32, [1]> var_5851_axes_0 = const()[name = tensor<string, []>("op_5851_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 1024, 600]> var_5851_cast_fp16 = squeeze(axes = var_5851_axes_0, x = x_5_cast_fp16)[name = tensor<string, []>("op_5851_cast_fp16")];
            tensor<int32, [3]> out_99_perm_0 = const()[name = tensor<string, []>("out_99_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> var_5867_pad_type_0 = const()[name = tensor<string, []>("op_5867_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_5867_pad_0 = const()[name = tensor<string, []>("op_5867_pad_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [1]> var_5867_strides_0 = const()[name = tensor<string, []>("op_5867_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> var_5867_dilations_0 = const()[name = tensor<string, []>("op_5867_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> var_5867_groups_0 = const()[name = tensor<string, []>("op_5867_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 1024, 3000]> var_5867_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_5867_dilations_0, groups = var_5867_groups_0, pad = var_5867_pad_0, pad_type = var_5867_pad_type_0, strides = var_5867_strides_0, weight = const_0_to_fp16_palettized, x = logmel_data)[name = tensor<string, []>("op_5867_cast_fp16")];
            tensor<string, []> input_245_mode_0 = const()[name = tensor<string, []>("input_245_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 3000]> input_245_cast_fp16 = gelu(mode = input_245_mode_0, x = var_5867_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")];
            tensor<string, []> var_5883_pad_type_0 = const()[name = tensor<string, []>("op_5883_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> var_5883_pad_0 = const()[name = tensor<string, []>("op_5883_pad_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [1]> var_5883_strides_0 = const()[name = tensor<string, []>("op_5883_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> var_5883_dilations_0 = const()[name = tensor<string, []>("op_5883_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> var_5883_groups_0 = const()[name = tensor<string, []>("op_5883_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 1024, 1500]> var_5883_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_5883_dilations_0, groups = var_5883_groups_0, pad = var_5883_pad_0, pad_type = var_5883_pad_type_0, strides = var_5883_strides_0, weight = const_2_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor<string, []>("op_5883_cast_fp16")];
            tensor<string, []> other_part_1_mode_0 = const()[name = tensor<string, []>("other_part_1_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1024, 1500]> other_part_1_cast_fp16 = gelu(mode = other_part_1_mode_0, x = var_5883_cast_fp16)[name = tensor<string, []>("other_part_1_cast_fp16")];
            tensor<int32, [3]> other_part_3_perm_0 = const()[name = tensor<string, []>("other_part_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> var_5910_begin_0 = const()[name = tensor<string, []>("op_5910_begin_0"), val = tensor<int32, [3]>([0, 600, 0])];
            tensor<int32, [3]> var_5910_end_0 = const()[name = tensor<string, []>("op_5910_end_0"), val = tensor<int32, [3]>([1, 1500, 1024])];
            tensor<bool, [3]> var_5910_end_mask_0 = const()[name = tensor<string, []>("op_5910_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 1500, 1024]> other_part_3_cast_fp16 = transpose(perm = other_part_3_perm_0, x = other_part_1_cast_fp16)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [1, 900, 1024]> var_5910_cast_fp16 = slice_by_index(begin = var_5910_begin_0, end = var_5910_end_0, end_mask = var_5910_end_mask_0, x = other_part_3_cast_fp16)[name = tensor<string, []>("op_5910_cast_fp16")];
            tensor<fp16, [1, 900, 1024]> op_5895_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153638272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154099136))), name = tensor<string, []>("op_5895_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 900, 1024])];
            tensor<fp16, [1, 900, 1024]> other_part_cast_fp16 = add(x = var_5910_cast_fp16, y = op_5895_to_fp16_palettized)[name = tensor<string, []>("other_part_cast_fp16")];
            tensor<int32, []> var_5919 = const()[name = tensor<string, []>("op_5919"), val = tensor<int32, []>(1)];
            tensor<bool, []> x_7_interleave_0 = const()[name = tensor<string, []>("x_7_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 600, 1024]> out_99_cast_fp16 = transpose(perm = out_99_perm_0, x = var_5851_cast_fp16)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [1, 1500, 1024]> x_7_cast_fp16 = concat(axis = var_5919, interleave = x_7_interleave_0, values = (out_99_cast_fp16, other_part_cast_fp16))[name = tensor<string, []>("x_7_cast_fp16")];
            tensor<int32, [1]> var_5943_axes_0 = const()[name = tensor<string, []>("op_5943_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> correction_block_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("correction_block_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154099264)))];
            tensor<fp16, [1024]> correction_block_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("correction_block_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154101376)))];
            tensor<fp16, []> var_5933_to_fp16 = const()[name = tensor<string, []>("op_5933_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 1500, 1024]> var_5943_cast_fp16 = layer_norm(axes = var_5943_axes_0, beta = correction_block_attn_ln_bias_to_fp16, epsilon = var_5933_to_fp16, gamma = correction_block_attn_ln_weight_to_fp16, x = x_7_cast_fp16)[name = tensor<string, []>("op_5943_cast_fp16")];
            tensor<fp16, [1024, 1024]> const_153_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154103488))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154627840))), name = tensor<string, []>("const_153_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])];
            tensor<fp16, [1024]> const_154_to_fp16 = const()[name = tensor<string, []>("const_154_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154627968)))];
            tensor<fp16, [1, 1500, 1024]> linear_0_cast_fp16 = linear(bias = const_154_to_fp16, weight = const_153_to_fp16_palettized, x = var_5943_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1024, 1024]> const_155_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154630080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155154432))), name = tensor<string, []>("const_155_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])];
            tensor<fp16, [1024]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155154560)))];
            tensor<fp16, [1, 1500, 1024]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_155_to_fp16_palettized, x = var_5943_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> const_156_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155156672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155681024))), name = tensor<string, []>("const_156_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])];
            tensor<fp16, [1024]> const_157_to_fp16 = const()[name = tensor<string, []>("const_157_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155681152)))];
            tensor<fp16, [1, 1500, 1024]> linear_2_cast_fp16 = linear(bias = const_157_to_fp16, weight = const_156_to_fp16_palettized, x = var_5943_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_5967 = const()[name = tensor<string, []>("op_5967"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
            tensor<fp16, [1, 1500, 16, 64]> var_5968_cast_fp16 = reshape(shape = var_5967, x = linear_0_cast_fp16)[name = tensor<string, []>("op_5968_cast_fp16")];
            tensor<int32, [4]> var_5973 = const()[name = tensor<string, []>("op_5973"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
            tensor<fp16, [1, 1500, 16, 64]> var_5974_cast_fp16 = reshape(shape = var_5973, x = linear_1_cast_fp16)[name = tensor<string, []>("op_5974_cast_fp16")];
            tensor<int32, [4]> var_5979 = const()[name = tensor<string, []>("op_5979"), val = tensor<int32, [4]>([1, 1500, 16, -1])];
            tensor<fp16, [1, 1500, 16, 64]> var_5980_cast_fp16 = reshape(shape = var_5979, x = linear_2_cast_fp16)[name = tensor<string, []>("op_5980_cast_fp16")];
            tensor<int32, [4]> var_5981 = const()[name = tensor<string, []>("op_5981"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> mul_144_y_0_to_fp16 = const()[name = tensor<string, []>("mul_144_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 1500, 16, 64]> mul_144_cast_fp16 = mul(x = var_5968_cast_fp16, y = mul_144_y_0_to_fp16)[name = tensor<string, []>("mul_144_cast_fp16")];
            tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)];
            tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = tensor<string, []>("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_5_perm_0 = const()[name = tensor<string, []>("transpose_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [1, 16, 1500, 64]> transpose_5 = transpose(perm = transpose_5_perm_0, x = var_5974_cast_fp16)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [1, 16, 1500, 64]> transpose_4 = transpose(perm = transpose_4_perm_0, x = mul_144_cast_fp16)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [1, 16, 1500, 1500]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_4, y = transpose_5)[name = tensor<string, []>("matmul_0_cast_fp16")];
            tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 1500, 1500]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = matmul_0_cast_fp16)[name = tensor<string, []>("softmax_0_cast_fp16")];
            tensor<bool, []> a_transpose_x_0 = const()[name = tensor<string, []>("a_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> a_transpose_y_0 = const()[name = tensor<string, []>("a_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1500, 64]> v_cast_fp16 = transpose(perm = var_5981, x = var_5980_cast_fp16)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [1, 16, 1500, 64]> a_cast_fp16 = matmul(transpose_x = a_transpose_x_0, transpose_y = a_transpose_y_0, x = softmax_0_cast_fp16, y = v_cast_fp16)[name = tensor<string, []>("a_cast_fp16")];
            tensor<int32, [4]> var_5984 = const()[name = tensor<string, []>("op_5984"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_0 = const()[name = tensor<string, []>("concat_0"), val = tensor<int32, [3]>([1, 1500, 1024])];
            tensor<fp16, [1, 1500, 16, 64]> var_5985_cast_fp16 = transpose(perm = var_5984, x = a_cast_fp16)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [1, 1500, 1024]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_5985_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
            tensor<fp16, [1024, 1024]> const_164_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [524288]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155683264))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156207616))), name = tensor<string, []>("const_164_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])];
            tensor<fp16, [1024]> const_165_to_fp16 = const()[name = tensor<string, []>("const_165_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156207744)))];
            tensor<fp16, [1, 1500, 1024]> linear_3_cast_fp16 = linear(bias = const_165_to_fp16, weight = const_164_to_fp16_palettized, x = x_11_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 1500, 1024]> x_13_cast_fp16 = add(x = x_7_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
            tensor<int32, [1]> var_5997_axes_0 = const()[name = tensor<string, []>("op_5997_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> correction_block_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("correction_block_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156209856)))];
            tensor<fp16, [1024]> correction_block_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("correction_block_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156211968)))];
            tensor<fp16, [1, 1500, 1024]> var_5997_cast_fp16 = layer_norm(axes = var_5997_axes_0, beta = correction_block_mlp_ln_bias_to_fp16, epsilon = var_5933_to_fp16, gamma = correction_block_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor<string, []>("op_5997_cast_fp16")];
            tensor<fp16, [4096, 1024]> const_166_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156214080))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158311296))), name = tensor<string, []>("const_166_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])];
            tensor<fp16, [4096]> const_167_to_fp16 = const()[name = tensor<string, []>("const_167_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158311424)))];
            tensor<fp16, [1, 1500, 4096]> linear_4_cast_fp16 = linear(bias = const_167_to_fp16, weight = const_166_to_fp16_palettized, x = var_5997_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> x_mode_0 = const()[name = tensor<string, []>("x_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [1, 1500, 4096]> x_cast_fp16 = gelu(mode = x_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("x_cast_fp16")];
            tensor<fp16, [1024, 4096]> const_168_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158319680))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160416896))), name = tensor<string, []>("const_168_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])];
            tensor<fp16, [1024]> const_169_to_fp16 = const()[name = tensor<string, []>("const_169_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160417024)))];
            tensor<fp16, [1, 1500, 1024]> linear_5_cast_fp16 = linear(bias = const_169_to_fp16, weight = const_168_to_fp16_palettized, x = x_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 1500, 1024]> output = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("op_6015_cast_fp16")];
        } -> (output);
}