Add checkpoints

.gitattributes

@@ -1,6 +1,7 @@
 *.7z filter=lfs diff=lfs merge=lfs -text
 *.arrow filter=lfs diff=lfs merge=lfs -text
 *.bin filter=lfs diff=lfs merge=lfs -text
+*.bin.* filter=lfs diff=lfs merge=lfs -text
 *.bz2 filter=lfs diff=lfs merge=lfs -text
 *.ftz filter=lfs diff=lfs merge=lfs -text
 *.gz filter=lfs diff=lfs merge=lfs -text
@@ -16,11 +17,10 @@
 *.pt filter=lfs diff=lfs merge=lfs -text
 *.pth filter=lfs diff=lfs merge=lfs -text
 *.rar filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text 
 *.tar.* filter=lfs diff=lfs merge=lfs -text
 *.tflite filter=lfs diff=lfs merge=lfs -text
 *.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
 *.xz filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,68 @@
+---
+language: ru
+datasets:
+- Common Voice 
+metrics:
+- wer
+tags:
+- audio
+- speech
+- wav2vec2
+- Russian-speech-corpus
+- automatic-speech-recognition
+- speech
+- PyTorch
+license: apache-2.0
+model-index:
+- name: Edresson Casanova Wav2vec2 Large 100k Voxpopuli fine-tuned with a single-speaker dataset in Russian 
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    metrics:
+       - name: Test Common Voice 7.0 WER
+         type: wer
+         value: 74.02
+---
+
+# Wav2vec2 Large 100k Voxpopuli fine-tuned with a single-speaker dataset in Russian 
+
+[Wav2vec2 Large 100k Voxpopuli](https://huggingface.co/facebook/wav2vec2-large-100k-voxpopuli) fine-tuned in Russian using a single-speaker dataset.
+
+
+
+# Use this model
+
+```python
+
+from transformers import AutoTokenizer, Wav2Vec2ForCTC
+  
+tokenizer = AutoTokenizer.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-russian")
+
+model = Wav2Vec2ForCTC.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-russian")
+```
+# Results
+For the results check the [article (Soon)]()
+
+# Example test with Common Voice Dataset
+
+
+```python
+dataset = load_dataset("common_voice", "ru", split="test", data_dir="./cv-corpus-7.0-2021-07-21")
+
+resampler = torchaudio.transforms.Resampl(orig_freq=48_000, new_freq=16_000)
+
+def map_to_array(batch):
+    speech, _ = torchaudio.load(batch["path"])
+    batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
+    batch["sampling_rate"] = resampler.new_freq
+    batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower().replace("’", "'")
+    return batch
+```
+
+```python
+ds = dataset.map(map_to_array)
+result = ds.map(map_to_pred, batched=True, batch_size=1, remove_columns=list(ds.features.keys()))
+print(wer.compute(predictions=result["predicted"], references=result["target"]))
+```
+

all_results.json

@@ -0,0 +1,23 @@
+{
+    "epoch": 133.99,
+    "eval_loss": 0.5939580202102661,
+    "eval_mem_cpu_alloc_delta": 84615168,
+    "eval_mem_cpu_peaked_delta": 163840,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 9191723520,
+    "eval_runtime": 26.6646,
+    "eval_samples": 500,
+    "eval_samples_per_second": 18.751,
+    "eval_wer": 0.5386466591166478,
+    "init_mem_cpu_alloc_delta": 3938103296,
+    "init_mem_cpu_peaked_delta": 805023744,
+    "init_mem_gpu_alloc_delta": 1261915136,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 2171449344,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778437120,
+    "train_mem_gpu_peaked_delta": 10009929728,
+    "train_runtime": 115622.8375,
+    "train_samples": 6805,
+    "train_samples_per_second": 0.042
+}

config.json

@@ -0,0 +1,84 @@
+{
+  "_name_or_path": "facebook/wav2vec2-large-100k-voxpopuli",
+  "activation_dropout": 0.0,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": true,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.1,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.05,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "proj_codevector_dim": 768,
+  "transformers_version": "4.6.1",
+  "vocab_size": 39
+}

config_train.json

@@ -0,0 +1,163 @@
+{
+    "run_name": "Wav2Vec-fine-tuning-TEDx",
+    "run_description": "Fine tuning TEDx",
+    "seed": 42,
+    // AUDIO PARAMS
+    "sampling_rate": 16000,
+
+    // VOCABULARY PARAMETERS
+    "vocab":{
+        "vocab_path": "example/vocab_example_ru.json", // generic vocab for Portuguese
+        "blank": "<pad>", // blank token for padding
+        "silence": "|", // token between words
+        "unk": "<unk>" // unk token
+    },
+
+    // TRAINING
+    "batch_size": 16,       // Batch size for training.
+    "mixed_precision": true,     // level of optimization with NVIDIA's apex feature for automatic mixed FP16/FP32 precision (AMP), NOTE: currently only O1 is supported, and use "O1" to activate.
+    "early_stop_epochs": 10, // If 0 disabled else Number of epochs for stop training with validation loss dont decrease 
+    "preprocess_dataset": false, // if true, the dataset will be pre-processed and saved in disk, otherwise the audio files will be loaded in each step. Preprocessing makes training faster, but requires much more disk space.
+
+    // OPTIMIZER
+    "epochs": 140,                // total number of epochs to train.
+    "lr": 0.00003,                  // Initial learning rate.
+    "gradient_accumulation_steps": 12, 
+
+    // LOGGING
+    "logging_steps": 100,    // Number of steps to plot.
+    "load_best_model_at_end": true,
+    "save_total_limit": 3,
+    "warmup_ratio": 0.04761904762142857, // 0 disable Ratio of total training steps used for a linear warmup from 0 to learning_rate
+    "warmup_steps": 0, // 0 disable  Number of steps used for a linear warmup from 0 to learning_rate
+
+    // DATA LOADING
+    "num_loader_workers": 4,        // number of training data loader processes. Don't set it too big. 4-8 are goo
+    
+    // MODEL
+    "freeze_feature_extractor": true, // Whether to freeze the feature extractor layers of the model.
+    "attention_dropout": 0.1, // The dropout ratio for the attention probabilities.
+    "activation_dropout": 0.1, // The dropout ratio for activations inside the fully connected layer.
+    "hidden_dropout": 0.1, // The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+    "feat_proj_dropout": 0.1, // The dropout probabilitiy for all 1D convolutional layers in feature extractor.
+    "mask_time_prob": 0.05, //  Propability of each feature vector along the time axis to be chosen as the start of the vector span to be masked.
+    "layerdrop": 0.0,  // The LayerDrop probability. 
+    "gradient_checkpointing": true,  // If True, use gradient checkpointing to save memory at the expense of slower backward pass.
+
+    // ToDo: Implement Time mask and Frequency Mask
+    "audio_augmentation":[
+        // additive noise and room impulse response (RIR) simulation similar to: https://arxiv.org/pdf/2009.14153.pdf
+        {
+            "name": "additive",
+            "sounds_path":"/raid/datasets/DA/musan/speech/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 13.0,
+            "max_snr_in_db": 20.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        
+        {
+            "name": "additive",
+            "sounds_path":"/raid/datasets/DA/musan/music/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 5.0,
+            "max_snr_in_db": 15.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        {
+            "name": "additive",
+            "sounds_path":"/raid/datasets/DA/musan/noise/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 0.0,
+            "max_snr_in_db": 15.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        // rir filter proposed by: https://ieeexplore.ieee.org/document/7953152
+        {
+            "name": "rir",
+            "ir_path": "/raid/datasets/DA/RIRS_NOISES/simulated_rirs/", // download: https://www.openslr.org/28/
+            "lru_cache_size": 128, // Maximum size of the LRU cache for storing noise files in memory
+            // "sample_rate": 16000,
+            "p": 0.25
+        }
+        , 
+        // {
+        //     "name": "gain",
+        //     "min_gain_in_db": -18.0,
+        //     "max_gain_in_db": 6,
+        //     "p": 0.25 // propability of apply this method, 0 is disable
+        // },
+        {
+            "name": "pitch_shift",
+            "min_semitones": -4,
+            "max_semitones": 4,
+            "p": 0.25 // propability of apply this method, 0 is disable
+        },
+        {
+            "name": "gaussian",
+            "min_amplitude": 0.0001,
+            "max_amplitude": 0.001,
+            "p": 0.25 // propability of apply this method, 0 is disable
+        }
+    ],
+
+    // PATHS
+    "output_path": "../checkpoints/YourTTS2ASR/Wav2Vec-voxpopuli/one-speaker/just-TTS/RU/140-epoch-high-bs/",
+    // CACHE
+    "dataset_cache": "../datasets/ru-cache-high-bs/",
+
+    // DATASETS
+    "datasets":{         
+        "files_path": "/raid/datasets/Mailabs/ru/", // relative path for audios It's will be join with the CS
+        "train":
+            [
+               // this dicts is pass directly for the load dataset see the documentation: https://huggingface.co/docs/datasets/package_reference/loading_methods.html#datasets.load_dataset 
+            {                
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/raid/datasets/Mailabs/ru/train_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }
+            ]
+        ,
+        "devel":
+            [ 
+                {                
+                    "name": "csv",
+                    "path": "csv",
+                    "data_files": ["/raid/datasets/Mailabs/ru/dev_converted.csv"], // csv files
+                    "text_column": "text",
+                    "path_column": "file_path"
+                }
+            ] 
+            ,
+        "test":
+            {
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/raid/datasets/DA/Common_Voice/cv-corpus-7.0-2021-07-21/ru/test_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }
+    
+    }//,
+    // used only for test 
+    // "KenLM":{
+    // "kenlm_model_path": "../../kenLM/binaries/subtitle/4-gram/lm.binary", // Path for KenLM model
+    // "lexicon_path": "example/lexicon.lst", // file with all words for limit the decoder search
+    // "beam": 2048, 
+    // "nbest": 1,
+    // "beam_threshold": 25, 
+    // "lm_weight": 1, 
+    // "word_score": -1,
+    // "sil_weight": 0
+    // }
+
+
+
+}
+

eval_results.json

@@ -0,0 +1,12 @@
+{
+    "epoch": 133.99,
+    "eval_loss": 0.5939580202102661,
+    "eval_mem_cpu_alloc_delta": 84615168,
+    "eval_mem_cpu_peaked_delta": 163840,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 9191723520,
+    "eval_runtime": 26.6646,
+    "eval_samples": 500,
+    "eval_samples_per_second": 18.751,
+    "eval_wer": 0.5386466591166478
+}

preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44e5226d711e2864f354aa456695746a9d041ccdc97af1bc4686429225400b57
+size 1262083569

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|"}

train_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 133.99,
+    "init_mem_cpu_alloc_delta": 3938103296,
+    "init_mem_cpu_peaked_delta": 805023744,
+    "init_mem_gpu_alloc_delta": 1261915136,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 2171449344,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778437120,
+    "train_mem_gpu_peaked_delta": 10009929728,
+    "train_runtime": 115622.8375,
+    "train_samples": 6805,
+    "train_samples_per_second": 0.042
+}

trainer_state.json

@@ -0,0 +1,1377 @@
+{
+  "best_metric": 0.5304960012435913,
+  "best_model_checkpoint": "../checkpoints/YourTTS2ASR/Wav2Vec-voxpopuli/one-speaker/just-TTS/RU/140-epoch-high-bs/checkpoint-4340",
+  "epoch": 133.98591549295776,
+  "global_step": 4690,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03,
+      "learning_rate": 1.282051282051282e-07,
+      "loss": 12.6527,
+      "step": 1
+    },
+    {
+      "epoch": 0.99,
+      "eval_loss": 17.671436309814453,
+      "eval_runtime": 26.4755,
+      "eval_samples_per_second": 18.885,
+      "eval_wer": 1.0060872027180068,
+      "step": 35
+    },
+    {
+      "epoch": 1.99,
+      "eval_loss": 15.852593421936035,
+      "eval_runtime": 26.5339,
+      "eval_samples_per_second": 18.844,
+      "eval_wer": 1.0018403171007928,
+      "step": 70
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1.2435897435897436e-05,
+      "loss": 17.6404,
+      "step": 100
+    },
+    {
+      "epoch": 2.99,
+      "eval_loss": 10.858827590942383,
+      "eval_runtime": 26.6038,
+      "eval_samples_per_second": 18.794,
+      "eval_wer": 1.0,
+      "step": 105
+    },
+    {
+      "epoch": 3.99,
+      "eval_loss": 7.959273338317871,
+      "eval_runtime": 26.4213,
+      "eval_samples_per_second": 18.924,
+      "eval_wer": 1.0,
+      "step": 140
+    },
+    {
+      "epoch": 4.99,
+      "eval_loss": 6.166073799133301,
+      "eval_runtime": 26.8692,
+      "eval_samples_per_second": 18.609,
+      "eval_wer": 1.0,
+      "step": 175
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 2.512820512820513e-05,
+      "loss": 9.2135,
+      "step": 200
+    },
+    {
+      "epoch": 5.99,
+      "eval_loss": 5.027560710906982,
+      "eval_runtime": 27.7308,
+      "eval_samples_per_second": 18.031,
+      "eval_wer": 1.0,
+      "step": 210
+    },
+    {
+      "epoch": 6.99,
+      "eval_loss": 4.288309574127197,
+      "eval_runtime": 26.9272,
+      "eval_samples_per_second": 18.569,
+      "eval_wer": 1.0,
+      "step": 245
+    },
+    {
+      "epoch": 7.99,
+      "eval_loss": 3.822075128555298,
+      "eval_runtime": 26.6029,
+      "eval_samples_per_second": 18.795,
+      "eval_wer": 1.0,
+      "step": 280
+    },
+    {
+      "epoch": 8.56,
+      "learning_rate": 2.960137162451779e-05,
+      "loss": 5.0261,
+      "step": 300
+    },
+    {
+      "epoch": 8.99,
+      "eval_loss": 3.5520131587982178,
+      "eval_runtime": 26.5759,
+      "eval_samples_per_second": 18.814,
+      "eval_wer": 1.0,
+      "step": 315
+    },
+    {
+      "epoch": 9.99,
+      "eval_loss": 3.379794120788574,
+      "eval_runtime": 27.8087,
+      "eval_samples_per_second": 17.98,
+      "eval_wer": 1.0,
+      "step": 350
+    },
+    {
+      "epoch": 10.99,
+      "eval_loss": 3.283963441848755,
+      "eval_runtime": 27.4577,
+      "eval_samples_per_second": 18.21,
+      "eval_wer": 1.0,
+      "step": 385
+    },
+    {
+      "epoch": 11.42,
+      "learning_rate": 2.8958422631804546e-05,
+      "loss": 3.6533,
+      "step": 400
+    },
+    {
+      "epoch": 11.99,
+      "eval_loss": 3.2323801517486572,
+      "eval_runtime": 27.1971,
+      "eval_samples_per_second": 18.384,
+      "eval_wer": 1.0,
+      "step": 420
+    },
+    {
+      "epoch": 12.99,
+      "eval_loss": 3.18682861328125,
+      "eval_runtime": 27.2137,
+      "eval_samples_per_second": 18.373,
+      "eval_wer": 1.0,
+      "step": 455
+    },
+    {
+      "epoch": 13.99,
+      "eval_loss": 3.1655502319335938,
+      "eval_runtime": 26.7209,
+      "eval_samples_per_second": 18.712,
+      "eval_wer": 1.0,
+      "step": 490
+    },
+    {
+      "epoch": 14.28,
+      "learning_rate": 2.83154736390913e-05,
+      "loss": 3.3493,
+      "step": 500
+    },
+    {
+      "epoch": 14.99,
+      "eval_loss": 3.1466615200042725,
+      "eval_runtime": 27.4367,
+      "eval_samples_per_second": 18.224,
+      "eval_wer": 1.0,
+      "step": 525
+    },
+    {
+      "epoch": 15.99,
+      "eval_loss": 3.1380531787872314,
+      "eval_runtime": 26.3265,
+      "eval_samples_per_second": 18.992,
+      "eval_wer": 1.0,
+      "step": 560
+    },
+    {
+      "epoch": 16.99,
+      "eval_loss": 3.136110305786133,
+      "eval_runtime": 26.5152,
+      "eval_samples_per_second": 18.857,
+      "eval_wer": 1.0,
+      "step": 595
+    },
+    {
+      "epoch": 17.14,
+      "learning_rate": 2.7672524646378054e-05,
+      "loss": 3.2398,
+      "step": 600
+    },
+    {
+      "epoch": 17.99,
+      "eval_loss": 3.107888698577881,
+      "eval_runtime": 27.5531,
+      "eval_samples_per_second": 18.147,
+      "eval_wer": 1.0,
+      "step": 630
+    },
+    {
+      "epoch": 18.99,
+      "eval_loss": 3.0973823070526123,
+      "eval_runtime": 26.2928,
+      "eval_samples_per_second": 19.017,
+      "eval_wer": 1.0,
+      "step": 665
+    },
+    {
+      "epoch": 19.99,
+      "learning_rate": 2.702957565366481e-05,
+      "loss": 3.1694,
+      "step": 700
+    },
+    {
+      "epoch": 19.99,
+      "eval_loss": 3.094352960586548,
+      "eval_runtime": 26.9913,
+      "eval_samples_per_second": 18.525,
+      "eval_wer": 1.0,
+      "step": 700
+    },
+    {
+      "epoch": 20.99,
+      "eval_loss": 3.057588577270508,
+      "eval_runtime": 26.8711,
+      "eval_samples_per_second": 18.607,
+      "eval_wer": 1.0,
+      "step": 735
+    },
+    {
+      "epoch": 21.99,
+      "eval_loss": 3.0422518253326416,
+      "eval_runtime": 26.285,
+      "eval_samples_per_second": 19.022,
+      "eval_wer": 1.0,
+      "step": 770
+    },
+    {
+      "epoch": 22.85,
+      "learning_rate": 2.6386626660951566e-05,
+      "loss": 3.1362,
+      "step": 800
+    },
+    {
+      "epoch": 22.99,
+      "eval_loss": 3.0276663303375244,
+      "eval_runtime": 26.5969,
+      "eval_samples_per_second": 18.799,
+      "eval_wer": 1.0,
+      "step": 805
+    },
+    {
+      "epoch": 23.99,
+      "eval_loss": 3.0015952587127686,
+      "eval_runtime": 26.954,
+      "eval_samples_per_second": 18.55,
+      "eval_wer": 1.0,
+      "step": 840
+    },
+    {
+      "epoch": 24.99,
+      "eval_loss": 2.9831387996673584,
+      "eval_runtime": 26.3639,
+      "eval_samples_per_second": 18.965,
+      "eval_wer": 1.0,
+      "step": 875
+    },
+    {
+      "epoch": 25.7,
+      "learning_rate": 2.574367766823832e-05,
+      "loss": 3.0785,
+      "step": 900
+    },
+    {
+      "epoch": 25.99,
+      "eval_loss": 2.9717295169830322,
+      "eval_runtime": 26.345,
+      "eval_samples_per_second": 18.979,
+      "eval_wer": 1.0,
+      "step": 910
+    },
+    {
+      "epoch": 26.99,
+      "eval_loss": 2.931703567504883,
+      "eval_runtime": 28.5603,
+      "eval_samples_per_second": 17.507,
+      "eval_wer": 1.0,
+      "step": 945
+    },
+    {
+      "epoch": 27.99,
+      "eval_loss": 2.907294511795044,
+      "eval_runtime": 27.6636,
+      "eval_samples_per_second": 18.074,
+      "eval_wer": 1.0,
+      "step": 980
+    },
+    {
+      "epoch": 28.56,
+      "learning_rate": 2.5100728675525077e-05,
+      "loss": 3.0127,
+      "step": 1000
+    },
+    {
+      "epoch": 28.99,
+      "eval_loss": 2.8216962814331055,
+      "eval_runtime": 27.1654,
+      "eval_samples_per_second": 18.406,
+      "eval_wer": 0.9988674971687429,
+      "step": 1015
+    },
+    {
+      "epoch": 29.99,
+      "eval_loss": 2.7435991764068604,
+      "eval_runtime": 27.0704,
+      "eval_samples_per_second": 18.47,
+      "eval_wer": 0.9941959229898075,
+      "step": 1050
+    },
+    {
+      "epoch": 30.99,
+      "eval_loss": 2.619396924972534,
+      "eval_runtime": 27.0157,
+      "eval_samples_per_second": 18.508,
+      "eval_wer": 0.9906568516421291,
+      "step": 1085
+    },
+    {
+      "epoch": 31.42,
+      "learning_rate": 2.4457779682811833e-05,
+      "loss": 2.8512,
+      "step": 1100
+    },
+    {
+      "epoch": 31.99,
+      "eval_loss": 2.5079009532928467,
+      "eval_runtime": 27.1587,
+      "eval_samples_per_second": 18.41,
+      "eval_wer": 0.9924971687429218,
+      "step": 1120
+    },
+    {
+      "epoch": 32.99,
+      "eval_loss": 2.3371741771698,
+      "eval_runtime": 27.7752,
+      "eval_samples_per_second": 18.002,
+      "eval_wer": 0.9917893544733862,
+      "step": 1155
+    },
+    {
+      "epoch": 33.99,
+      "eval_loss": 2.1688764095306396,
+      "eval_runtime": 27.3286,
+      "eval_samples_per_second": 18.296,
+      "eval_wer": 0.9882502831257078,
+      "step": 1190
+    },
+    {
+      "epoch": 34.28,
+      "learning_rate": 2.3814830690098586e-05,
+      "loss": 2.5708,
+      "step": 1200
+    },
+    {
+      "epoch": 34.99,
+      "eval_loss": 1.9715479612350464,
+      "eval_runtime": 26.5008,
+      "eval_samples_per_second": 18.867,
+      "eval_wer": 0.9685730464326161,
+      "step": 1225
+    },
+    {
+      "epoch": 35.99,
+      "eval_loss": 1.8692930936813354,
+      "eval_runtime": 26.8592,
+      "eval_samples_per_second": 18.616,
+      "eval_wer": 0.9617780294450736,
+      "step": 1260
+    },
+    {
+      "epoch": 36.99,
+      "eval_loss": 1.7158366441726685,
+      "eval_runtime": 26.8153,
+      "eval_samples_per_second": 18.646,
+      "eval_wer": 0.9658833522083805,
+      "step": 1295
+    },
+    {
+      "epoch": 37.14,
+      "learning_rate": 2.317188169738534e-05,
+      "loss": 2.1632,
+      "step": 1300
+    },
+    {
+      "epoch": 37.99,
+      "eval_loss": 1.5786515474319458,
+      "eval_runtime": 27.1788,
+      "eval_samples_per_second": 18.397,
+      "eval_wer": 0.943233295583239,
+      "step": 1330
+    },
+    {
+      "epoch": 38.99,
+      "eval_loss": 1.490448236465454,
+      "eval_runtime": 27.5548,
+      "eval_samples_per_second": 18.146,
+      "eval_wer": 0.9457814269535674,
+      "step": 1365
+    },
+    {
+      "epoch": 39.99,
+      "learning_rate": 2.2528932704672097e-05,
+      "loss": 1.8294,
+      "step": 1400
+    },
+    {
+      "epoch": 39.99,
+      "eval_loss": 1.4414215087890625,
+      "eval_runtime": 27.3752,
+      "eval_samples_per_second": 18.265,
+      "eval_wer": 0.9232729331823329,
+      "step": 1400
+    },
+    {
+      "epoch": 40.99,
+      "eval_loss": 1.3827379941940308,
+      "eval_runtime": 26.825,
+      "eval_samples_per_second": 18.639,
+      "eval_wer": 0.9113816534541337,
+      "step": 1435
+    },
+    {
+      "epoch": 41.99,
+      "eval_loss": 1.3141157627105713,
+      "eval_runtime": 26.9732,
+      "eval_samples_per_second": 18.537,
+      "eval_wer": 0.9089750849377124,
+      "step": 1470
+    },
+    {
+      "epoch": 42.85,
+      "learning_rate": 2.1885983711958853e-05,
+      "loss": 1.593,
+      "step": 1500
+    },
+    {
+      "epoch": 42.99,
+      "eval_loss": 1.2413103580474854,
+      "eval_runtime": 26.9193,
+      "eval_samples_per_second": 18.574,
+      "eval_wer": 0.8898640996602492,
+      "step": 1505
+    },
+    {
+      "epoch": 43.99,
+      "eval_loss": 1.2450193166732788,
+      "eval_runtime": 27.4051,
+      "eval_samples_per_second": 18.245,
+      "eval_wer": 0.878114382785957,
+      "step": 1540
+    },
+    {
+      "epoch": 44.99,
+      "eval_loss": 1.1904444694519043,
+      "eval_runtime": 27.0345,
+      "eval_samples_per_second": 18.495,
+      "eval_wer": 0.862400906002265,
+      "step": 1575
+    },
+    {
+      "epoch": 45.7,
+      "learning_rate": 2.124303471924561e-05,
+      "loss": 1.4381,
+      "step": 1600
+    },
+    {
+      "epoch": 45.99,
+      "eval_loss": 1.1529375314712524,
+      "eval_runtime": 27.9547,
+      "eval_samples_per_second": 17.886,
+      "eval_wer": 0.8573046432616082,
+      "step": 1610
+    },
+    {
+      "epoch": 46.99,
+      "eval_loss": 1.0873721837997437,
+      "eval_runtime": 27.3563,
+      "eval_samples_per_second": 18.277,
+      "eval_wer": 0.8425821064552661,
+      "step": 1645
+    },
+    {
+      "epoch": 47.99,
+      "eval_loss": 1.0788930654525757,
+      "eval_runtime": 27.3167,
+      "eval_samples_per_second": 18.304,
+      "eval_wer": 0.8338052095130238,
+      "step": 1680
+    },
+    {
+      "epoch": 48.56,
+      "learning_rate": 2.060008572653236e-05,
+      "loss": 1.3129,
+      "step": 1700
+    },
+    {
+      "epoch": 48.99,
+      "eval_loss": 1.0390797853469849,
+      "eval_runtime": 27.5562,
+      "eval_samples_per_second": 18.145,
+      "eval_wer": 0.8220554926387316,
+      "step": 1715
+    },
+    {
+      "epoch": 49.99,
+      "eval_loss": 0.9944789409637451,
+      "eval_runtime": 26.1901,
+      "eval_samples_per_second": 19.091,
+      "eval_wer": 0.8062004530011325,
+      "step": 1750
+    },
+    {
+      "epoch": 50.99,
+      "eval_loss": 0.9924176931381226,
+      "eval_runtime": 28.2948,
+      "eval_samples_per_second": 17.671,
+      "eval_wer": 0.8042185730464326,
+      "step": 1785
+    },
+    {
+      "epoch": 51.42,
+      "learning_rate": 1.9957136733819117e-05,
+      "loss": 1.2179,
+      "step": 1800
+    },
+    {
+      "epoch": 51.99,
+      "eval_loss": 0.9647215604782104,
+      "eval_runtime": 27.2768,
+      "eval_samples_per_second": 18.331,
+      "eval_wer": 0.7977066817667045,
+      "step": 1820
+    },
+    {
+      "epoch": 52.99,
+      "eval_loss": 0.959790050983429,
+      "eval_runtime": 27.5769,
+      "eval_samples_per_second": 18.131,
+      "eval_wer": 0.7843997734994338,
+      "step": 1855
+    },
+    {
+      "epoch": 53.99,
+      "eval_loss": 0.9579805135726929,
+      "eval_runtime": 26.7511,
+      "eval_samples_per_second": 18.691,
+      "eval_wer": 0.7768969422423556,
+      "step": 1890
+    },
+    {
+      "epoch": 54.28,
+      "learning_rate": 1.9314187741105873e-05,
+      "loss": 1.138,
+      "step": 1900
+    },
+    {
+      "epoch": 54.99,
+      "eval_loss": 0.9402398467063904,
+      "eval_runtime": 27.2724,
+      "eval_samples_per_second": 18.334,
+      "eval_wer": 0.7920441676104191,
+      "step": 1925
+    },
+    {
+      "epoch": 55.99,
+      "eval_loss": 0.8787918090820312,
+      "eval_runtime": 27.3961,
+      "eval_samples_per_second": 18.251,
+      "eval_wer": 0.7559456398640997,
+      "step": 1960
+    },
+    {
+      "epoch": 56.99,
+      "eval_loss": 0.8727829456329346,
+      "eval_runtime": 27.2979,
+      "eval_samples_per_second": 18.316,
+      "eval_wer": 0.7539637599093998,
+      "step": 1995
+    },
+    {
+      "epoch": 57.14,
+      "learning_rate": 1.867123874839263e-05,
+      "loss": 1.0772,
+      "step": 2000
+    },
+    {
+      "epoch": 57.99,
+      "eval_loss": 0.8611069321632385,
+      "eval_runtime": 28.307,
+      "eval_samples_per_second": 17.663,
+      "eval_wer": 0.7456115515288788,
+      "step": 2030
+    },
+    {
+      "epoch": 58.99,
+      "eval_loss": 0.8685981631278992,
+      "eval_runtime": 26.4664,
+      "eval_samples_per_second": 18.892,
+      "eval_wer": 0.7393827859569649,
+      "step": 2065
+    },
+    {
+      "epoch": 59.99,
+      "learning_rate": 1.8028289755679385e-05,
+      "loss": 1.0328,
+      "step": 2100
+    },
+    {
+      "epoch": 59.99,
+      "eval_loss": 0.8199361562728882,
+      "eval_runtime": 26.9722,
+      "eval_samples_per_second": 18.538,
+      "eval_wer": 0.7260758776896942,
+      "step": 2100
+    },
+    {
+      "epoch": 60.99,
+      "eval_loss": 0.8023450374603271,
+      "eval_runtime": 27.0456,
+      "eval_samples_per_second": 18.487,
+      "eval_wer": 0.7250849377123443,
+      "step": 2135
+    },
+    {
+      "epoch": 61.99,
+      "eval_loss": 0.8279299139976501,
+      "eval_runtime": 27.3268,
+      "eval_samples_per_second": 18.297,
+      "eval_wer": 0.7280577576443941,
+      "step": 2170
+    },
+    {
+      "epoch": 62.85,
+      "learning_rate": 1.7385340762966137e-05,
+      "loss": 0.9874,
+      "step": 2200
+    },
+    {
+      "epoch": 62.99,
+      "eval_loss": 0.8266852498054504,
+      "eval_runtime": 26.5957,
+      "eval_samples_per_second": 18.8,
+      "eval_wer": 0.7056908267270668,
+      "step": 2205
+    },
+    {
+      "epoch": 63.99,
+      "eval_loss": 0.7818687558174133,
+      "eval_runtime": 27.0982,
+      "eval_samples_per_second": 18.451,
+      "eval_wer": 0.7042751981879954,
+      "step": 2240
+    },
+    {
+      "epoch": 64.99,
+      "eval_loss": 0.8026483058929443,
+      "eval_runtime": 27.0042,
+      "eval_samples_per_second": 18.516,
+      "eval_wer": 0.7004530011325029,
+      "step": 2275
+    },
+    {
+      "epoch": 65.7,
+      "learning_rate": 1.6742391770252893e-05,
+      "loss": 0.949,
+      "step": 2300
+    },
+    {
+      "epoch": 65.99,
+      "eval_loss": 0.8021445274353027,
+      "eval_runtime": 27.2917,
+      "eval_samples_per_second": 18.321,
+      "eval_wer": 0.6971970554926388,
+      "step": 2310
+    },
+    {
+      "epoch": 66.99,
+      "eval_loss": 0.7785004377365112,
+      "eval_runtime": 28.0805,
+      "eval_samples_per_second": 17.806,
+      "eval_wer": 0.6956398640996603,
+      "step": 2345
+    },
+    {
+      "epoch": 67.99,
+      "eval_loss": 0.7500312328338623,
+      "eval_runtime": 26.9245,
+      "eval_samples_per_second": 18.57,
+      "eval_wer": 0.6820498301245753,
+      "step": 2380
+    },
+    {
+      "epoch": 68.56,
+      "learning_rate": 1.609944277753965e-05,
+      "loss": 0.9119,
+      "step": 2400
+    },
+    {
+      "epoch": 68.99,
+      "eval_loss": 0.701343834400177,
+      "eval_runtime": 26.782,
+      "eval_samples_per_second": 18.669,
+      "eval_wer": 0.6768120045300113,
+      "step": 2415
+    },
+    {
+      "epoch": 69.99,
+      "eval_loss": 0.7393462061882019,
+      "eval_runtime": 27.5438,
+      "eval_samples_per_second": 18.153,
+      "eval_wer": 0.6719988674971688,
+      "step": 2450
+    },
+    {
+      "epoch": 70.99,
+      "eval_loss": 0.7068197131156921,
+      "eval_runtime": 27.1804,
+      "eval_samples_per_second": 18.396,
+      "eval_wer": 0.6626557191392979,
+      "step": 2485
+    },
+    {
+      "epoch": 71.42,
+      "learning_rate": 1.5456493784826405e-05,
+      "loss": 0.876,
+      "step": 2500
+    },
+    {
+      "epoch": 71.99,
+      "eval_loss": 0.7229353189468384,
+      "eval_runtime": 25.5356,
+      "eval_samples_per_second": 19.581,
+      "eval_wer": 0.6599660249150623,
+      "step": 2520
+    },
+    {
+      "epoch": 72.99,
+      "eval_loss": 0.7125120759010315,
+      "eval_runtime": 26.7798,
+      "eval_samples_per_second": 18.671,
+      "eval_wer": 0.6550113250283126,
+      "step": 2555
+    },
+    {
+      "epoch": 73.99,
+      "eval_loss": 0.6882209181785583,
+      "eval_runtime": 27.3156,
+      "eval_samples_per_second": 18.305,
+      "eval_wer": 0.6520385050962627,
+      "step": 2590
+    },
+    {
+      "epoch": 74.28,
+      "learning_rate": 1.4813544792113159e-05,
+      "loss": 0.8639,
+      "step": 2600
+    },
+    {
+      "epoch": 74.99,
+      "eval_loss": 0.6938254237174988,
+      "eval_runtime": 27.0308,
+      "eval_samples_per_second": 18.497,
+      "eval_wer": 0.6394394110985278,
+      "step": 2625
+    },
+    {
+      "epoch": 75.99,
+      "eval_loss": 0.7523351311683655,
+      "eval_runtime": 26.9197,
+      "eval_samples_per_second": 18.574,
+      "eval_wer": 0.6540203850509626,
+      "step": 2660
+    },
+    {
+      "epoch": 76.99,
+      "eval_loss": 0.6973133683204651,
+      "eval_runtime": 27.134,
+      "eval_samples_per_second": 18.427,
+      "eval_wer": 0.6507644394110985,
+      "step": 2695
+    },
+    {
+      "epoch": 77.14,
+      "learning_rate": 1.4170595799399915e-05,
+      "loss": 0.8319,
+      "step": 2700
+    },
+    {
+      "epoch": 77.99,
+      "eval_loss": 0.6753961443901062,
+      "eval_runtime": 27.0856,
+      "eval_samples_per_second": 18.46,
+      "eval_wer": 0.621885617214043,
+      "step": 2730
+    },
+    {
+      "epoch": 78.99,
+      "eval_loss": 0.6487002372741699,
+      "eval_runtime": 27.3912,
+      "eval_samples_per_second": 18.254,
+      "eval_wer": 0.6213193657984145,
+      "step": 2765
+    },
+    {
+      "epoch": 79.99,
+      "learning_rate": 1.352764680668667e-05,
+      "loss": 0.8096,
+      "step": 2800
+    },
+    {
+      "epoch": 79.99,
+      "eval_loss": 0.6611467003822327,
+      "eval_runtime": 27.457,
+      "eval_samples_per_second": 18.21,
+      "eval_wer": 0.6288221970554927,
+      "step": 2800
+    },
+    {
+      "epoch": 80.99,
+      "eval_loss": 0.667143702507019,
+      "eval_runtime": 27.4086,
+      "eval_samples_per_second": 18.242,
+      "eval_wer": 0.6104190260475651,
+      "step": 2835
+    },
+    {
+      "epoch": 81.99,
+      "eval_loss": 0.6765517592430115,
+      "eval_runtime": 26.975,
+      "eval_samples_per_second": 18.536,
+      "eval_wer": 0.616647791619479,
+      "step": 2870
+    },
+    {
+      "epoch": 82.85,
+      "learning_rate": 1.2884697813973425e-05,
+      "loss": 0.7862,
+      "step": 2900
+    },
+    {
+      "epoch": 82.99,
+      "eval_loss": 0.6575422286987305,
+      "eval_runtime": 26.7138,
+      "eval_samples_per_second": 18.717,
+      "eval_wer": 0.6099943374858438,
+      "step": 2905
+    },
+    {
+      "epoch": 83.99,
+      "eval_loss": 0.6632807850837708,
+      "eval_runtime": 26.7198,
+      "eval_samples_per_second": 18.713,
+      "eval_wer": 0.6109852774631936,
+      "step": 2940
+    },
+    {
+      "epoch": 84.99,
+      "eval_loss": 0.6666624546051025,
+      "eval_runtime": 27.1802,
+      "eval_samples_per_second": 18.396,
+      "eval_wer": 0.616647791619479,
+      "step": 2975
+    },
+    {
+      "epoch": 85.7,
+      "learning_rate": 1.224174882126018e-05,
+      "loss": 0.7795,
+      "step": 3000
+    },
+    {
+      "epoch": 85.99,
+      "eval_loss": 0.6229019165039062,
+      "eval_runtime": 27.3639,
+      "eval_samples_per_second": 18.272,
+      "eval_wer": 0.5975368063420159,
+      "step": 3010
+    },
+    {
+      "epoch": 86.99,
+      "eval_loss": 0.6241843104362488,
+      "eval_runtime": 27.6099,
+      "eval_samples_per_second": 18.109,
+      "eval_wer": 0.5881936579841449,
+      "step": 3045
+    },
+    {
+      "epoch": 87.99,
+      "eval_loss": 0.6389002799987793,
+      "eval_runtime": 27.1816,
+      "eval_samples_per_second": 18.395,
+      "eval_wer": 0.5962627406568517,
+      "step": 3080
+    },
+    {
+      "epoch": 88.56,
+      "learning_rate": 1.1598799828546935e-05,
+      "loss": 0.76,
+      "step": 3100
+    },
+    {
+      "epoch": 88.99,
+      "eval_loss": 0.6407724618911743,
+      "eval_runtime": 27.9119,
+      "eval_samples_per_second": 17.914,
+      "eval_wer": 0.5982446206115515,
+      "step": 3115
+    },
+    {
+      "epoch": 89.99,
+      "eval_loss": 0.6397743821144104,
+      "eval_runtime": 26.1239,
+      "eval_samples_per_second": 19.14,
+      "eval_wer": 0.5846545866364666,
+      "step": 3150
+    },
+    {
+      "epoch": 90.99,
+      "eval_loss": 0.6100246906280518,
+      "eval_runtime": 26.8354,
+      "eval_samples_per_second": 18.632,
+      "eval_wer": 0.5849377123442808,
+      "step": 3185
+    },
+    {
+      "epoch": 91.42,
+      "learning_rate": 1.095585083583369e-05,
+      "loss": 0.74,
+      "step": 3200
+    },
+    {
+      "epoch": 91.99,
+      "eval_loss": 0.6202873587608337,
+      "eval_runtime": 27.7182,
+      "eval_samples_per_second": 18.039,
+      "eval_wer": 0.5818233295583239,
+      "step": 3220
+    },
+    {
+      "epoch": 92.99,
+      "eval_loss": 0.609228253364563,
+      "eval_runtime": 27.272,
+      "eval_samples_per_second": 18.334,
+      "eval_wer": 0.5784258210645526,
+      "step": 3255
+    },
+    {
+      "epoch": 93.99,
+      "eval_loss": 0.6129232048988342,
+      "eval_runtime": 27.5404,
+      "eval_samples_per_second": 18.155,
+      "eval_wer": 0.57559456398641,
+      "step": 3290
+    },
+    {
+      "epoch": 94.28,
+      "learning_rate": 1.0312901843120446e-05,
+      "loss": 0.7256,
+      "step": 3300
+    },
+    {
+      "epoch": 94.99,
+      "eval_loss": 0.6256955862045288,
+      "eval_runtime": 26.854,
+      "eval_samples_per_second": 18.619,
+      "eval_wer": 0.5791336353340883,
+      "step": 3325
+    },
+    {
+      "epoch": 95.99,
+      "eval_loss": 0.623586893081665,
+      "eval_runtime": 26.9076,
+      "eval_samples_per_second": 18.582,
+      "eval_wer": 0.5741789354473387,
+      "step": 3360
+    },
+    {
+      "epoch": 96.99,
+      "eval_loss": 0.6311513781547546,
+      "eval_runtime": 26.8392,
+      "eval_samples_per_second": 18.629,
+      "eval_wer": 0.5716308040770102,
+      "step": 3395
+    },
+    {
+      "epoch": 97.14,
+      "learning_rate": 9.6699528504072e-06,
+      "loss": 0.7228,
+      "step": 3400
+    },
+    {
+      "epoch": 97.99,
+      "eval_loss": 0.6243217587471008,
+      "eval_runtime": 26.966,
+      "eval_samples_per_second": 18.542,
+      "eval_wer": 0.584088335220838,
+      "step": 3430
+    },
+    {
+      "epoch": 98.99,
+      "eval_loss": 0.6131792664527893,
+      "eval_runtime": 26.9072,
+      "eval_samples_per_second": 18.582,
+      "eval_wer": 0.5697904869762175,
+      "step": 3465
+    },
+    {
+      "epoch": 99.99,
+      "learning_rate": 9.027003857693956e-06,
+      "loss": 0.7052,
+      "step": 3500
+    },
+    {
+      "epoch": 99.99,
+      "eval_loss": 0.6121107339859009,
+      "eval_runtime": 27.0528,
+      "eval_samples_per_second": 18.482,
+      "eval_wer": 0.5740373725934315,
+      "step": 3500
+    },
+    {
+      "epoch": 100.99,
+      "eval_loss": 0.5875396728515625,
+      "eval_runtime": 26.9794,
+      "eval_samples_per_second": 18.533,
+      "eval_wer": 0.5571913929784824,
+      "step": 3535
+    },
+    {
+      "epoch": 101.99,
+      "eval_loss": 0.6091192364692688,
+      "eval_runtime": 27.0699,
+      "eval_samples_per_second": 18.471,
+      "eval_wer": 0.5780011325028312,
+      "step": 3570
+    },
+    {
+      "epoch": 102.85,
+      "learning_rate": 8.384054864980712e-06,
+      "loss": 0.7004,
+      "step": 3600
+    },
+    {
+      "epoch": 102.99,
+      "eval_loss": 0.5911608934402466,
+      "eval_runtime": 26.9942,
+      "eval_samples_per_second": 18.523,
+      "eval_wer": 0.5560588901472253,
+      "step": 3605
+    },
+    {
+      "epoch": 103.99,
+      "eval_loss": 0.6172874569892883,
+      "eval_runtime": 27.4328,
+      "eval_samples_per_second": 18.226,
+      "eval_wer": 0.5686579841449604,
+      "step": 3640
+    },
+    {
+      "epoch": 104.99,
+      "eval_loss": 0.5960651636123657,
+      "eval_runtime": 25.9346,
+      "eval_samples_per_second": 19.279,
+      "eval_wer": 0.5518120045300113,
+      "step": 3675
+    },
+    {
+      "epoch": 105.7,
+      "learning_rate": 7.741105872267466e-06,
+      "loss": 0.6821,
+      "step": 3700
+    },
+    {
+      "epoch": 105.99,
+      "eval_loss": 0.6121740937232971,
+      "eval_runtime": 26.1378,
+      "eval_samples_per_second": 19.129,
+      "eval_wer": 0.5595979614949037,
+      "step": 3710
+    },
+    {
+      "epoch": 106.99,
+      "eval_loss": 0.5817015171051025,
+      "eval_runtime": 26.989,
+      "eval_samples_per_second": 18.526,
+      "eval_wer": 0.5535107587768969,
+      "step": 3745
+    },
+    {
+      "epoch": 107.99,
+      "eval_loss": 0.579744279384613,
+      "eval_runtime": 26.8303,
+      "eval_samples_per_second": 18.636,
+      "eval_wer": 0.5373725934314836,
+      "step": 3780
+    },
+    {
+      "epoch": 108.56,
+      "learning_rate": 7.098156879554223e-06,
+      "loss": 0.6829,
+      "step": 3800
+    },
+    {
+      "epoch": 108.99,
+      "eval_loss": 0.6141317486763,
+      "eval_runtime": 26.7877,
+      "eval_samples_per_second": 18.665,
+      "eval_wer": 0.5554926387315968,
+      "step": 3815
+    },
+    {
+      "epoch": 109.99,
+      "eval_loss": 0.5983940362930298,
+      "eval_runtime": 26.6721,
+      "eval_samples_per_second": 18.746,
+      "eval_wer": 0.5532276330690826,
+      "step": 3850
+    },
+    {
+      "epoch": 110.99,
+      "eval_loss": 0.5953497290611267,
+      "eval_runtime": 27.2912,
+      "eval_samples_per_second": 18.321,
+      "eval_wer": 0.5607304643261608,
+      "step": 3885
+    },
+    {
+      "epoch": 111.42,
+      "learning_rate": 6.455207886840978e-06,
+      "loss": 0.6706,
+      "step": 3900
+    },
+    {
+      "epoch": 111.99,
+      "eval_loss": 0.5864156484603882,
+      "eval_runtime": 27.4851,
+      "eval_samples_per_second": 18.192,
+      "eval_wer": 0.5526613816534541,
+      "step": 3920
+    },
+    {
+      "epoch": 112.99,
+      "eval_loss": 0.589518666267395,
+      "eval_runtime": 26.5061,
+      "eval_samples_per_second": 18.864,
+      "eval_wer": 0.5244903737259343,
+      "step": 3955
+    },
+    {
+      "epoch": 113.99,
+      "eval_loss": 0.6098873615264893,
+      "eval_runtime": 26.9359,
+      "eval_samples_per_second": 18.563,
+      "eval_wer": 0.5562004530011325,
+      "step": 3990
+    },
+    {
+      "epoch": 114.28,
+      "learning_rate": 5.812258894127733e-06,
+      "loss": 0.6676,
+      "step": 4000
+    },
+    {
+      "epoch": 114.99,
+      "eval_loss": 0.593589186668396,
+      "eval_runtime": 27.4328,
+      "eval_samples_per_second": 18.226,
+      "eval_wer": 0.5383635334088335,
+      "step": 4025
+    },
+    {
+      "epoch": 115.99,
+      "eval_loss": 0.5409448146820068,
+      "eval_runtime": 26.749,
+      "eval_samples_per_second": 18.692,
+      "eval_wer": 0.5396375990939978,
+      "step": 4060
+    },
+    {
+      "epoch": 116.99,
+      "eval_loss": 0.6060774922370911,
+      "eval_runtime": 27.4597,
+      "eval_samples_per_second": 18.208,
+      "eval_wer": 0.5600226500566251,
+      "step": 4095
+    },
+    {
+      "epoch": 117.14,
+      "learning_rate": 5.169309901414488e-06,
+      "loss": 0.6627,
+      "step": 4100
+    },
+    {
+      "epoch": 117.99,
+      "eval_loss": 0.5925233364105225,
+      "eval_runtime": 27.1637,
+      "eval_samples_per_second": 18.407,
+      "eval_wer": 0.5481313703284258,
+      "step": 4130
+    },
+    {
+      "epoch": 118.99,
+      "eval_loss": 0.5753343105316162,
+      "eval_runtime": 26.9095,
+      "eval_samples_per_second": 18.581,
+      "eval_wer": 0.5465741789354474,
+      "step": 4165
+    },
+    {
+      "epoch": 119.99,
+      "learning_rate": 4.526360908701243e-06,
+      "loss": 0.6513,
+      "step": 4200
+    },
+    {
+      "epoch": 119.99,
+      "eval_loss": 0.5831518769264221,
+      "eval_runtime": 27.2243,
+      "eval_samples_per_second": 18.366,
+      "eval_wer": 0.5393544733861835,
+      "step": 4200
+    },
+    {
+      "epoch": 120.99,
+      "eval_loss": 0.6026310920715332,
+      "eval_runtime": 27.2645,
+      "eval_samples_per_second": 18.339,
+      "eval_wer": 0.5423272933182333,
+      "step": 4235
+    },
+    {
+      "epoch": 121.99,
+      "eval_loss": 0.6404977440834045,
+      "eval_runtime": 28.1762,
+      "eval_samples_per_second": 17.745,
+      "eval_wer": 0.5603057757644394,
+      "step": 4270
+    },
+    {
+      "epoch": 122.85,
+      "learning_rate": 3.883411915987999e-06,
+      "loss": 0.6537,
+      "step": 4300
+    },
+    {
+      "epoch": 122.99,
+      "eval_loss": 0.5665517449378967,
+      "eval_runtime": 27.5119,
+      "eval_samples_per_second": 18.174,
+      "eval_wer": 0.527321630804077,
+      "step": 4305
+    },
+    {
+      "epoch": 123.99,
+      "eval_loss": 0.5304960012435913,
+      "eval_runtime": 27.367,
+      "eval_samples_per_second": 18.27,
+      "eval_wer": 0.5196772366930917,
+      "step": 4340
+    },
+    {
+      "epoch": 124.99,
+      "eval_loss": 0.6143291592597961,
+      "eval_runtime": 27.7739,
+      "eval_samples_per_second": 18.002,
+      "eval_wer": 0.5349660249150623,
+      "step": 4375
+    },
+    {
+      "epoch": 125.7,
+      "learning_rate": 3.2404629232747538e-06,
+      "loss": 0.6465,
+      "step": 4400
+    },
+    {
+      "epoch": 125.99,
+      "eval_loss": 0.5990394353866577,
+      "eval_runtime": 27.6536,
+      "eval_samples_per_second": 18.081,
+      "eval_wer": 0.5335503963759909,
+      "step": 4410
+    },
+    {
+      "epoch": 126.99,
+      "eval_loss": 0.5602393746376038,
+      "eval_runtime": 26.8983,
+      "eval_samples_per_second": 18.589,
+      "eval_wer": 0.5257644394110985,
+      "step": 4445
+    },
+    {
+      "epoch": 127.99,
+      "eval_loss": 0.6085386276245117,
+      "eval_runtime": 27.9177,
+      "eval_samples_per_second": 17.91,
+      "eval_wer": 0.5562004530011325,
+      "step": 4480
+    },
+    {
+      "epoch": 128.56,
+      "learning_rate": 2.5975139305615088e-06,
+      "loss": 0.6407,
+      "step": 4500
+    },
+    {
+      "epoch": 128.99,
+      "eval_loss": 0.613398551940918,
+      "eval_runtime": 27.1482,
+      "eval_samples_per_second": 18.417,
+      "eval_wer": 0.530436013590034,
+      "step": 4515
+    },
+    {
+      "epoch": 129.99,
+      "eval_loss": 0.5771151185035706,
+      "eval_runtime": 26.8929,
+      "eval_samples_per_second": 18.592,
+      "eval_wer": 0.5293035107587769,
+      "step": 4550
+    },
+    {
+      "epoch": 130.99,
+      "eval_loss": 0.5841774940490723,
+      "eval_runtime": 26.9872,
+      "eval_samples_per_second": 18.527,
+      "eval_wer": 0.5310022650056625,
+      "step": 4585
+    },
+    {
+      "epoch": 131.42,
+      "learning_rate": 1.954564937848264e-06,
+      "loss": 0.6354,
+      "step": 4600
+    },
+    {
+      "epoch": 131.99,
+      "eval_loss": 0.6040024161338806,
+      "eval_runtime": 27.4951,
+      "eval_samples_per_second": 18.185,
+      "eval_wer": 0.5479898074745186,
+      "step": 4620
+    },
+    {
+      "epoch": 132.99,
+      "eval_loss": 0.5754969716072083,
+      "eval_runtime": 27.655,
+      "eval_samples_per_second": 18.08,
+      "eval_wer": 0.5185447338618346,
+      "step": 4655
+    },
+    {
+      "epoch": 133.99,
+      "eval_loss": 0.5896986126899719,
+      "eval_runtime": 25.767,
+      "eval_samples_per_second": 19.405,
+      "eval_wer": 0.5288788221970555,
+      "step": 4690
+    },
+    {
+      "epoch": 133.99,
+      "step": 4690,
+      "total_flos": 0,
+      "train_runtime": 115622.8375,
+      "train_samples_per_second": 0.042
+    }
+  ],
+  "max_steps": 4900,
+  "num_train_epochs": 140,
+  "total_flos": 0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5776e8549de6ea4bd7b48698a6b7e41ea19e156c560c75f26a23cc31407aa50d
+size 2607

vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "|": 4, "-": 5, "ё": 6, "а": 7, "б": 8, "в": 9, "г": 10, "д": 11, "е": 12, "ж": 13, "з": 14, "и": 15, "й": 16, "к": 17, "л": 18, "м": 19, "н": 20, "о": 21, "п": 22, "р": 23, "с": 24, "т": 25, "у": 26, "ф": 27, "х": 28, "ц": 29, "ч": 30, "ш": 31, "щ": 32, "ъ": 33, "ы": 34, "ь": 35, "э": 36, "ю": 37, "я": 38}