fix

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "pszemraj_nougat-small-onnx",
+  "_name_or_path": "pszemraj_nougat-base-onnx",
   "architectures": [
     "VisionEncoderDecoderModel"
   ],
@@ -21,7 +21,7 @@
     "decoder_attention_heads": 16,
     "decoder_ffn_dim": 4096,
     "decoder_layerdrop": 0.0,
-    "decoder_layers": 4,
+    "decoder_layers": 10,
     "decoder_start_token_id": null,
     "diversity_penalty": 0.0,
     "do_sample": false,
@@ -50,7 +50,7 @@
     },
     "length_penalty": 1.0,
     "max_length": 20,
-    "max_position_embeddings": 3584,
+    "max_position_embeddings": 4096,
     "min_length": 0,
     "model_type": "mbart",
     "no_repeat_ngram_size": 0,

decoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9c8f1bb7456441320245595f93fa052612b9f477b71f4822a46c070faf48e8d
-size 174444105
+oid sha256:15da4726c8d041f961d9b97481effca9ddba49446df6dc3f17eeec64740aa348
+size 277026563

decoder_with_past_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a68ddb99f8e476a6ba4e356145440de0fa70a5d2cff89c136feba9691f27c98
-size 165923718
+oid sha256:5381e07724b3177510b22fb04dfbaf6378740b4f73bf4822868e1d7b300a06d3
+size 255737635

encoder_model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6bec906871770afe111f0f3474bd5cb143cfd7f659c242431ee5431be580684d
-size 82052587
+oid sha256:4966de105b16f2208c41b6993b02f2ab36dbfefd7de897200f11c0781b4c9c33
+size 82052674

ort_config.json

@@ -2,7 +2,7 @@
   "one_external_file": true,
   "opset": null,
   "optimization": {},
-  "optimum_version": "1.17.1",
+  "optimum_version": "1.18.0.dev0",
   "quantization": {
     "activations_dtype": "QUInt8",
     "activations_symmetric": false,

tokenizer.json

@@ -2,13 +2,13 @@
   "version": "1.0",
   "truncation": {
     "direction": "Right",
-    "max_length": 3584,
+    "max_length": 4096,
     "strategy": "LongestFirst",
     "stride": 0
   },
   "padding": {
     "strategy": {
-      "Fixed": 3584
+      "Fixed": 4096
     },
     "direction": "Right",
     "pad_to_multiple_of": null,

tokenizer_config.json

@@ -189,7 +189,7 @@
   "bos_token": "<s>",
   "clean_up_tokenization_spaces": false,
   "eos_token": "</s>",
-  "max_length": 3584,
+  "max_length": 4096,
   "model_max_length": 3584,
   "pad_to_multiple_of": null,
   "pad_token": "<pad>",