Add checkpoints

README.md

@@ -0,0 +1,69 @@
+---
+language: pt
+datasets:
+- Common Voice 
+metrics:
+- wer
+tags:
+- audio
+- speech
+- wav2vec2
+- pt
+- 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 plus Data Augmentation in Russian 
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    metrics:
+       - name: Test Common Voice 7.0 WER
+         type: wer
+         value: 36.59
+---
+
+# Wav2vec2 Large 100k Voxpopuli fine-tuned with a single-speaker dataset plus Data Augmentation in Russian 
+
+[Wav2vec2 Large 100k Voxpopuli](https://huggingface.co/facebook/wav2vec2-large-100k-voxpopuli) fine-tuned in Russian using a single-speaker dataset plus a data augmentation method based on TTS and voice conversion.
+
+
+
+# Use this model
+
+```python
+
+from transformers import AutoTokenizer, Wav2Vec2ForCTC
+  
+tokenizer = AutoTokenizer.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-plus-data-augmentation-russian")
+
+model = Wav2Vec2ForCTC.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-plus-data-augmentation-russian")
+```
+# Results
+For the results check the [article (Soon)]()
+
+# Example test with Common Voice Dataset
+
+
+```python
+dataset = load_dataset("common_voice", "pt", split="test", data_dir="./cv-corpus-7.0-2021-07-21")
+
+resampler = torchaudio.transforms.Resample(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": 58.0,
+    "eval_loss": 0.3337676227092743,
+    "eval_mem_cpu_alloc_delta": 258572288,
+    "eval_mem_cpu_peaked_delta": 458752,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 3631953920,
+    "eval_runtime": 106.6624,
+    "eval_samples": 1500,
+    "eval_samples_per_second": 14.063,
+    "eval_wer": 0.2851510469391017,
+    "init_mem_cpu_alloc_delta": 1956712448,
+    "init_mem_cpu_peaked_delta": 0,
+    "init_mem_gpu_alloc_delta": 1261915136,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 1570873344,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778065408,
+    "train_mem_gpu_peaked_delta": 4841915392,
+    "train_runtime": 265934.449,
+    "train_samples": 57900,
+    "train_samples_per_second": 0.113
+}

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 +1,191 @@
+{
+    "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": 8,       // 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": 100,                // total number of epochs to train.
+    "lr": 0.00003,                  // Initial learning rate.
+    "gradient_accumulation_steps": 24, 
+
+    // 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": 8,        // 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":"/workspace/scratch/ecasanova/datasets/YourTTS2ASR/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":"/workspace/scratch/ecasanova/datasets/YourTTS2ASR/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":"/workspace/scratch/ecasanova/datasets/YourTTS2ASR/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": "/workspace/scratch/ecasanova/datasets/YourTTS2ASR/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/Wav2Vec-voxpopuli/one-speaker/Final-paper/GEN/RU/100-epoch/",
+    // CACHE
+    "dataset_cache": "../datasets/",
+
+    // DATASETS
+    "datasets":{ 
+        
+        "files_path": "/workspace/scratch/ecasanova/datasets/YourTTS2ASR/Common_Voice/cv-corpus-7.0-2021-07-21/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": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/Common_Voice/cv-corpus-7.0-2021-07-21/ru/train_converted_copy_generated_en_speakers.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }, 
+            {                
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/M-AILABS/ru_RU/train_converted_copy_generated_VC_en_speakers_5_speakers_per_text_fixed.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            },
+            {                
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/M-AILABS/ru_RU/train_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }
+            ]
+        ,
+        "devel":
+            [ 
+                {
+                    "name": "csv",
+                    "path": "csv",
+                    "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/Common_Voice/cv-corpus-7.0-2021-07-21/ru/dev_converted_copy_generated_en_speakers_500-samples.csv"], // csv files
+                    "text_column": "text",
+                    "path_column": "file_path"
+                }, 
+                {                
+                    "name": "csv",
+                    "path": "csv",
+                    "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/M-AILABS/ru_RU/dev_converted.csv"], // csv files
+                    "text_column": "text",
+                    "path_column": "file_path"
+                }, 
+                {                
+                    "name": "csv",
+                    "path": "csv",
+                    "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/M-AILABS/ru_RU/dev_converted_VC_5_speakers_500-samples.csv"], // csv files
+                    "text_column": "text",
+                    "path_column": "file_path"
+                }
+            ] 
+            ,
+        "test":
+            {
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/scratch/ecasanova/datasets/YourTTS2ASR/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": 58.0,
+    "eval_loss": 0.3337676227092743,
+    "eval_mem_cpu_alloc_delta": 258572288,
+    "eval_mem_cpu_peaked_delta": 458752,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 3631953920,
+    "eval_runtime": 106.6624,
+    "eval_samples": 1500,
+    "eval_samples_per_second": 14.063,
+    "eval_wer": 0.2851510469391017
+}

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:b3ea64738060fe41294b9ceb930cc53929dfd1a18174aab46f2f3474f76053e8
+size 1262093719

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": 58.0,
+    "init_mem_cpu_alloc_delta": 1956712448,
+    "init_mem_cpu_peaked_delta": 0,
+    "init_mem_gpu_alloc_delta": 1261915136,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 1570873344,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778065408,
+    "train_mem_gpu_peaked_delta": 4841915392,
+    "train_runtime": 265934.449,
+    "train_samples": 57900,
+    "train_samples_per_second": 0.113
+}

trainer_state.json

@@ -0,0 +1,1537 @@
+{
+  "best_metric": 0.32980483770370483,
+  "best_model_checkpoint": "../checkpoints/Wav2Vec-voxpopuli/one-speaker/Final-paper/GEN/RU/100-epoch/checkpoint-14448",
+  "epoch": 57.99806576402321,
+  "global_step": 17458,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.092050209205021e-08,
+      "loss": 12.2973,
+      "step": 1
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.0502092050209206e-06,
+      "loss": 16.675,
+      "step": 100
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 4.1422594142259415e-06,
+      "loss": 14.7926,
+      "step": 200
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 6.213389121338913e-06,
+      "loss": 10.0497,
+      "step": 300
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 7.385106086730957,
+      "eval_runtime": 109.638,
+      "eval_samples_per_second": 13.681,
+      "eval_wer": 1.0,
+      "step": 301
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 8.305439330543932e-06,
+      "loss": 7.5239,
+      "step": 400
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.0397489539748954e-05,
+      "loss": 5.8939,
+      "step": 500
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.2489539748953975e-05,
+      "loss": 4.7804,
+      "step": 600
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 3.9045944213867188,
+      "eval_runtime": 109.7364,
+      "eval_samples_per_second": 13.669,
+      "eval_wer": 1.0,
+      "step": 602
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 1.4581589958158997e-05,
+      "loss": 4.0957,
+      "step": 700
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 1.667364016736402e-05,
+      "loss": 3.5995,
+      "step": 800
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 1.8765690376569037e-05,
+      "loss": 3.3599,
+      "step": 900
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 3.2242395877838135,
+      "eval_runtime": 109.2767,
+      "eval_samples_per_second": 13.727,
+      "eval_wer": 1.0,
+      "step": 903
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 2.085774058577406e-05,
+      "loss": 3.2604,
+      "step": 1000
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 2.294979079497908e-05,
+      "loss": 3.167,
+      "step": 1100
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 2.50418410041841e-05,
+      "loss": 3.1242,
+      "step": 1200
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 3.069446563720703,
+      "eval_runtime": 109.1339,
+      "eval_samples_per_second": 13.745,
+      "eval_wer": 1.0,
+      "step": 1204
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 2.713389121338912e-05,
+      "loss": 3.0867,
+      "step": 1300
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 2.9225941422594142e-05,
+      "loss": 2.9571,
+      "step": 1400
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 2.9934068234144983e-05,
+      "loss": 2.7874,
+      "step": 1500
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 2.4769504070281982,
+      "eval_runtime": 109.8357,
+      "eval_samples_per_second": 13.657,
+      "eval_wer": 0.9946720656401513,
+      "step": 1505
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 2.982941463754971e-05,
+      "loss": 2.425,
+      "step": 1600
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 2.972476104095444e-05,
+      "loss": 2.0164,
+      "step": 1700
+    },
+    {
+      "epoch": 5.98,
+      "learning_rate": 2.9620107444359173e-05,
+      "loss": 1.7913,
+      "step": 1800
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 1.4069615602493286,
+      "eval_runtime": 109.5694,
+      "eval_samples_per_second": 13.69,
+      "eval_wer": 0.9387287548617401,
+      "step": 1806
+    },
+    {
+      "epoch": 6.31,
+      "learning_rate": 2.95154538477639e-05,
+      "loss": 1.5987,
+      "step": 1900
+    },
+    {
+      "epoch": 6.64,
+      "learning_rate": 2.9410800251168633e-05,
+      "loss": 1.4588,
+      "step": 2000
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 2.9306146654573363e-05,
+      "loss": 1.3903,
+      "step": 2100
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 1.077704668045044,
+      "eval_runtime": 109.7039,
+      "eval_samples_per_second": 13.673,
+      "eval_wer": 0.8479407533699185,
+      "step": 2107
+    },
+    {
+      "epoch": 7.31,
+      "learning_rate": 2.9201493057978096e-05,
+      "loss": 1.2888,
+      "step": 2200
+    },
+    {
+      "epoch": 7.64,
+      "learning_rate": 2.9096839461382823e-05,
+      "loss": 1.2117,
+      "step": 2300
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 2.8992185864787553e-05,
+      "loss": 1.1329,
+      "step": 2400
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 0.8992915153503418,
+      "eval_runtime": 109.3945,
+      "eval_samples_per_second": 13.712,
+      "eval_wer": 0.7582183387500666,
+      "step": 2408
+    },
+    {
+      "epoch": 8.31,
+      "learning_rate": 2.8887532268192286e-05,
+      "loss": 1.0772,
+      "step": 2500
+    },
+    {
+      "epoch": 8.64,
+      "learning_rate": 2.8782878671597013e-05,
+      "loss": 1.0268,
+      "step": 2600
+    },
+    {
+      "epoch": 8.97,
+      "learning_rate": 2.8678225075001746e-05,
+      "loss": 0.9772,
+      "step": 2700
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 0.7786251902580261,
+      "eval_runtime": 110.3302,
+      "eval_samples_per_second": 13.596,
+      "eval_wer": 0.694123288401087,
+      "step": 2709
+    },
+    {
+      "epoch": 9.3,
+      "learning_rate": 2.8573571478406476e-05,
+      "loss": 0.9467,
+      "step": 2800
+    },
+    {
+      "epoch": 9.63,
+      "learning_rate": 2.8468917881811206e-05,
+      "loss": 0.9015,
+      "step": 2900
+    },
+    {
+      "epoch": 9.96,
+      "learning_rate": 2.8364264285215936e-05,
+      "loss": 0.865,
+      "step": 3000
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 0.6784546971321106,
+      "eval_runtime": 108.8544,
+      "eval_samples_per_second": 13.78,
+      "eval_wer": 0.6233150407586978,
+      "step": 3010
+    },
+    {
+      "epoch": 10.3,
+      "learning_rate": 2.8259610688620666e-05,
+      "loss": 0.8455,
+      "step": 3100
+    },
+    {
+      "epoch": 10.63,
+      "learning_rate": 2.8154957092025396e-05,
+      "loss": 0.8152,
+      "step": 3200
+    },
+    {
+      "epoch": 10.96,
+      "learning_rate": 2.8050303495430126e-05,
+      "loss": 0.7937,
+      "step": 3300
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 0.6180887222290039,
+      "eval_runtime": 108.7247,
+      "eval_samples_per_second": 13.796,
+      "eval_wer": 0.5741382066172944,
+      "step": 3311
+    },
+    {
+      "epoch": 11.3,
+      "learning_rate": 2.794564989883486e-05,
+      "loss": 0.772,
+      "step": 3400
+    },
+    {
+      "epoch": 11.63,
+      "learning_rate": 2.7840996302239586e-05,
+      "loss": 0.7565,
+      "step": 3500
+    },
+    {
+      "epoch": 11.96,
+      "learning_rate": 2.7736342705644316e-05,
+      "loss": 0.7299,
+      "step": 3600
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 0.5893837809562683,
+      "eval_runtime": 108.6254,
+      "eval_samples_per_second": 13.809,
+      "eval_wer": 0.5368426660983536,
+      "step": 3612
+    },
+    {
+      "epoch": 12.29,
+      "learning_rate": 2.763168910904905e-05,
+      "loss": 0.7253,
+      "step": 3700
+    },
+    {
+      "epoch": 12.62,
+      "learning_rate": 2.7527035512453776e-05,
+      "loss": 0.7002,
+      "step": 3800
+    },
+    {
+      "epoch": 12.95,
+      "learning_rate": 2.742238191585851e-05,
+      "loss": 0.6854,
+      "step": 3900
+    },
+    {
+      "epoch": 13.0,
+      "eval_loss": 0.5517454147338867,
+      "eval_runtime": 108.415,
+      "eval_samples_per_second": 13.836,
+      "eval_wer": 0.5131866375406255,
+      "step": 3913
+    },
+    {
+      "epoch": 13.29,
+      "learning_rate": 2.731772831926324e-05,
+      "loss": 0.6758,
+      "step": 4000
+    },
+    {
+      "epoch": 13.62,
+      "learning_rate": 2.7213074722667973e-05,
+      "loss": 0.6529,
+      "step": 4100
+    },
+    {
+      "epoch": 13.95,
+      "learning_rate": 2.71084211260727e-05,
+      "loss": 0.6428,
+      "step": 4200
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 0.5316258668899536,
+      "eval_runtime": 107.9497,
+      "eval_samples_per_second": 13.895,
+      "eval_wer": 0.4877723906441473,
+      "step": 4214
+    },
+    {
+      "epoch": 14.29,
+      "learning_rate": 2.700376752947743e-05,
+      "loss": 0.6402,
+      "step": 4300
+    },
+    {
+      "epoch": 14.62,
+      "learning_rate": 2.6899113932882163e-05,
+      "loss": 0.626,
+      "step": 4400
+    },
+    {
+      "epoch": 14.95,
+      "learning_rate": 2.679446033628689e-05,
+      "loss": 0.6054,
+      "step": 4500
+    },
+    {
+      "epoch": 15.0,
+      "eval_loss": 0.5172051787376404,
+      "eval_runtime": 110.1009,
+      "eval_samples_per_second": 13.624,
+      "eval_wer": 0.476690287175662,
+      "step": 4515
+    },
+    {
+      "epoch": 15.28,
+      "learning_rate": 2.6689806739691623e-05,
+      "loss": 0.6079,
+      "step": 4600
+    },
+    {
+      "epoch": 15.61,
+      "learning_rate": 2.6585153143096353e-05,
+      "loss": 0.5898,
+      "step": 4700
+    },
+    {
+      "epoch": 15.95,
+      "learning_rate": 2.6480499546501082e-05,
+      "loss": 0.5854,
+      "step": 4800
+    },
+    {
+      "epoch": 16.0,
+      "eval_loss": 0.4909590780735016,
+      "eval_runtime": 109.8916,
+      "eval_samples_per_second": 13.65,
+      "eval_wer": 0.4592679418189568,
+      "step": 4816
+    },
+    {
+      "epoch": 16.28,
+      "learning_rate": 2.6375845949905812e-05,
+      "loss": 0.5883,
+      "step": 4900
+    },
+    {
+      "epoch": 16.61,
+      "learning_rate": 2.6271192353310542e-05,
+      "loss": 0.5702,
+      "step": 5000
+    },
+    {
+      "epoch": 16.94,
+      "learning_rate": 2.6166538756715272e-05,
+      "loss": 0.5608,
+      "step": 5100
+    },
+    {
+      "epoch": 17.0,
+      "eval_loss": 0.46681666374206543,
+      "eval_runtime": 109.9917,
+      "eval_samples_per_second": 13.637,
+      "eval_wer": 0.4439767702061911,
+      "step": 5117
+    },
+    {
+      "epoch": 17.28,
+      "learning_rate": 2.6061885160120002e-05,
+      "loss": 0.5568,
+      "step": 5200
+    },
+    {
+      "epoch": 17.61,
+      "learning_rate": 2.5957231563524736e-05,
+      "loss": 0.5499,
+      "step": 5300
+    },
+    {
+      "epoch": 17.94,
+      "learning_rate": 2.5852577966929462e-05,
+      "loss": 0.5364,
+      "step": 5400
+    },
+    {
+      "epoch": 18.0,
+      "eval_loss": 0.45651692152023315,
+      "eval_runtime": 109.4038,
+      "eval_samples_per_second": 13.711,
+      "eval_wer": 0.41861580265331133,
+      "step": 5418
+    },
+    {
+      "epoch": 18.27,
+      "learning_rate": 2.5747924370334196e-05,
+      "loss": 0.5324,
+      "step": 5500
+    },
+    {
+      "epoch": 18.6,
+      "learning_rate": 2.5643270773738926e-05,
+      "loss": 0.5174,
+      "step": 5600
+    },
+    {
+      "epoch": 18.94,
+      "learning_rate": 2.5538617177143652e-05,
+      "loss": 0.5139,
+      "step": 5700
+    },
+    {
+      "epoch": 19.0,
+      "eval_loss": 0.46214792132377625,
+      "eval_runtime": 109.5407,
+      "eval_samples_per_second": 13.694,
+      "eval_wer": 0.41765677446853855,
+      "step": 5719
+    },
+    {
+      "epoch": 19.27,
+      "learning_rate": 2.5433963580548386e-05,
+      "loss": 0.5139,
+      "step": 5800
+    },
+    {
+      "epoch": 19.6,
+      "learning_rate": 2.5329309983953116e-05,
+      "loss": 0.507,
+      "step": 5900
+    },
+    {
+      "epoch": 19.93,
+      "learning_rate": 2.522465638735785e-05,
+      "loss": 0.497,
+      "step": 6000
+    },
+    {
+      "epoch": 20.0,
+      "eval_loss": 0.43304443359375,
+      "eval_runtime": 108.5948,
+      "eval_samples_per_second": 13.813,
+      "eval_wer": 0.4080132132772124,
+      "step": 6020
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 2.5120002790762576e-05,
+      "loss": 0.4927,
+      "step": 6100
+    },
+    {
+      "epoch": 20.6,
+      "learning_rate": 2.501534919416731e-05,
+      "loss": 0.4899,
+      "step": 6200
+    },
+    {
+      "epoch": 20.93,
+      "learning_rate": 2.491069559757204e-05,
+      "loss": 0.4814,
+      "step": 6300
+    },
+    {
+      "epoch": 21.0,
+      "eval_loss": 0.4360824525356293,
+      "eval_runtime": 110.0234,
+      "eval_samples_per_second": 13.633,
+      "eval_wer": 0.39538600884437103,
+      "step": 6321
+    },
+    {
+      "epoch": 21.26,
+      "learning_rate": 2.4806042000976765e-05,
+      "loss": 0.485,
+      "step": 6400
+    },
+    {
+      "epoch": 21.59,
+      "learning_rate": 2.47013884043815e-05,
+      "loss": 0.4758,
+      "step": 6500
+    },
+    {
+      "epoch": 21.93,
+      "learning_rate": 2.459673480778623e-05,
+      "loss": 0.4701,
+      "step": 6600
+    },
+    {
+      "epoch": 22.0,
+      "eval_loss": 0.4039258360862732,
+      "eval_runtime": 109.5914,
+      "eval_samples_per_second": 13.687,
+      "eval_wer": 0.38233256966274176,
+      "step": 6622
+    },
+    {
+      "epoch": 22.26,
+      "learning_rate": 2.449208121119096e-05,
+      "loss": 0.4679,
+      "step": 6700
+    },
+    {
+      "epoch": 22.59,
+      "learning_rate": 2.438742761459569e-05,
+      "loss": 0.4601,
+      "step": 6800
+    },
+    {
+      "epoch": 22.92,
+      "learning_rate": 2.428277401800042e-05,
+      "loss": 0.4598,
+      "step": 6900
+    },
+    {
+      "epoch": 23.0,
+      "eval_loss": 0.41757893562316895,
+      "eval_runtime": 111.0731,
+      "eval_samples_per_second": 13.505,
+      "eval_wer": 0.3865949171506207,
+      "step": 6923
+    },
+    {
+      "epoch": 23.26,
+      "learning_rate": 2.417812042140515e-05,
+      "loss": 0.4511,
+      "step": 7000
+    },
+    {
+      "epoch": 23.59,
+      "learning_rate": 2.407346682480988e-05,
+      "loss": 0.4478,
+      "step": 7100
+    },
+    {
+      "epoch": 23.92,
+      "learning_rate": 2.3968813228214612e-05,
+      "loss": 0.4403,
+      "step": 7200
+    },
+    {
+      "epoch": 24.0,
+      "eval_loss": 0.40201762318611145,
+      "eval_runtime": 110.2327,
+      "eval_samples_per_second": 13.608,
+      "eval_wer": 0.37631200383611274,
+      "step": 7224
+    },
+    {
+      "epoch": 24.25,
+      "learning_rate": 2.386415963161934e-05,
+      "loss": 0.4447,
+      "step": 7300
+    },
+    {
+      "epoch": 24.58,
+      "learning_rate": 2.3759506035024072e-05,
+      "loss": 0.4318,
+      "step": 7400
+    },
+    {
+      "epoch": 24.92,
+      "learning_rate": 2.3654852438428802e-05,
+      "loss": 0.4335,
+      "step": 7500
+    },
+    {
+      "epoch": 25.0,
+      "eval_loss": 0.4044724404811859,
+      "eval_runtime": 109.3689,
+      "eval_samples_per_second": 13.715,
+      "eval_wer": 0.3676274708295594,
+      "step": 7525
+    },
+    {
+      "epoch": 25.25,
+      "learning_rate": 2.355019884183353e-05,
+      "loss": 0.4301,
+      "step": 7600
+    },
+    {
+      "epoch": 25.58,
+      "learning_rate": 2.3445545245238262e-05,
+      "loss": 0.4259,
+      "step": 7700
+    },
+    {
+      "epoch": 25.91,
+      "learning_rate": 2.3340891648642992e-05,
+      "loss": 0.4167,
+      "step": 7800
+    },
+    {
+      "epoch": 26.0,
+      "eval_loss": 0.3963695466518402,
+      "eval_runtime": 109.9175,
+      "eval_samples_per_second": 13.647,
+      "eval_wer": 0.35324204805796794,
+      "step": 7826
+    },
+    {
+      "epoch": 26.25,
+      "learning_rate": 2.3236238052047725e-05,
+      "loss": 0.4211,
+      "step": 7900
+    },
+    {
+      "epoch": 26.58,
+      "learning_rate": 2.3131584455452452e-05,
+      "loss": 0.4162,
+      "step": 8000
+    },
+    {
+      "epoch": 26.91,
+      "learning_rate": 2.3026930858857185e-05,
+      "loss": 0.4073,
+      "step": 8100
+    },
+    {
+      "epoch": 27.0,
+      "eval_loss": 0.3953684866428375,
+      "eval_runtime": 109.8131,
+      "eval_samples_per_second": 13.66,
+      "eval_wer": 0.37252917044062017,
+      "step": 8127
+    },
+    {
+      "epoch": 27.24,
+      "learning_rate": 2.2922277262261915e-05,
+      "loss": 0.4095,
+      "step": 8200
+    },
+    {
+      "epoch": 27.57,
+      "learning_rate": 2.281762366566664e-05,
+      "loss": 0.3996,
+      "step": 8300
+    },
+    {
+      "epoch": 27.91,
+      "learning_rate": 2.2712970069071375e-05,
+      "loss": 0.4001,
+      "step": 8400
+    },
+    {
+      "epoch": 28.0,
+      "eval_loss": 0.3944612145423889,
+      "eval_runtime": 111.2731,
+      "eval_samples_per_second": 13.48,
+      "eval_wer": 0.35809046832543023,
+      "step": 8428
+    },
+    {
+      "epoch": 28.24,
+      "learning_rate": 2.2608316472476105e-05,
+      "loss": 0.4028,
+      "step": 8500
+    },
+    {
+      "epoch": 28.57,
+      "learning_rate": 2.2503662875880835e-05,
+      "loss": 0.3949,
+      "step": 8600
+    },
+    {
+      "epoch": 28.9,
+      "learning_rate": 2.2399009279285565e-05,
+      "loss": 0.3891,
+      "step": 8700
+    },
+    {
+      "epoch": 29.0,
+      "eval_loss": 0.36043721437454224,
+      "eval_runtime": 110.8305,
+      "eval_samples_per_second": 13.534,
+      "eval_wer": 0.3408279609995205,
+      "step": 8729
+    },
+    {
+      "epoch": 29.24,
+      "learning_rate": 2.2294355682690298e-05,
+      "loss": 0.3911,
+      "step": 8800
+    },
+    {
+      "epoch": 29.57,
+      "learning_rate": 2.2189702086095025e-05,
+      "loss": 0.3849,
+      "step": 8900
+    },
+    {
+      "epoch": 29.9,
+      "learning_rate": 2.2085048489499755e-05,
+      "loss": 0.3794,
+      "step": 9000
+    },
+    {
+      "epoch": 30.0,
+      "eval_loss": 0.3697544038295746,
+      "eval_runtime": 110.2616,
+      "eval_samples_per_second": 13.604,
+      "eval_wer": 0.3352869092652779,
+      "step": 9030
+    },
+    {
+      "epoch": 30.23,
+      "learning_rate": 2.1980394892904488e-05,
+      "loss": 0.3793,
+      "step": 9100
+    },
+    {
+      "epoch": 30.56,
+      "learning_rate": 2.1875741296309215e-05,
+      "loss": 0.379,
+      "step": 9200
+    },
+    {
+      "epoch": 30.9,
+      "learning_rate": 2.1771087699713948e-05,
+      "loss": 0.3768,
+      "step": 9300
+    },
+    {
+      "epoch": 31.0,
+      "eval_loss": 0.3718980848789215,
+      "eval_runtime": 110.5675,
+      "eval_samples_per_second": 13.566,
+      "eval_wer": 0.33129095849539136,
+      "step": 9331
+    },
+    {
+      "epoch": 31.23,
+      "learning_rate": 2.1666434103118678e-05,
+      "loss": 0.3755,
+      "step": 9400
+    },
+    {
+      "epoch": 31.56,
+      "learning_rate": 2.1561780506523408e-05,
+      "loss": 0.3668,
+      "step": 9500
+    },
+    {
+      "epoch": 31.89,
+      "learning_rate": 2.1457126909928138e-05,
+      "loss": 0.3701,
+      "step": 9600
+    },
+    {
+      "epoch": 32.0,
+      "eval_loss": 0.37526383996009827,
+      "eval_runtime": 109.4741,
+      "eval_samples_per_second": 13.702,
+      "eval_wer": 0.32884010868986097,
+      "step": 9632
+    },
+    {
+      "epoch": 32.23,
+      "learning_rate": 2.1352473313332868e-05,
+      "loss": 0.3678,
+      "step": 9700
+    },
+    {
+      "epoch": 32.56,
+      "learning_rate": 2.12478197167376e-05,
+      "loss": 0.3649,
+      "step": 9800
+    },
+    {
+      "epoch": 32.89,
+      "learning_rate": 2.1143166120142328e-05,
+      "loss": 0.3677,
+      "step": 9900
+    },
+    {
+      "epoch": 33.0,
+      "eval_loss": 0.3619440495967865,
+      "eval_runtime": 110.6886,
+      "eval_samples_per_second": 13.552,
+      "eval_wer": 0.3304384889978155,
+      "step": 9933
+    },
+    {
+      "epoch": 33.22,
+      "learning_rate": 2.103851252354706e-05,
+      "loss": 0.3575,
+      "step": 10000
+    },
+    {
+      "epoch": 33.55,
+      "learning_rate": 2.093385892695179e-05,
+      "loss": 0.3593,
+      "step": 10100
+    },
+    {
+      "epoch": 33.89,
+      "learning_rate": 2.083025186632247e-05,
+      "loss": 0.3536,
+      "step": 10200
+    },
+    {
+      "epoch": 34.0,
+      "eval_loss": 0.36355456709861755,
+      "eval_runtime": 110.1201,
+      "eval_samples_per_second": 13.621,
+      "eval_wer": 0.322499866801641,
+      "step": 10234
+    },
+    {
+      "epoch": 34.22,
+      "learning_rate": 2.0725598269727205e-05,
+      "loss": 0.3493,
+      "step": 10300
+    },
+    {
+      "epoch": 34.55,
+      "learning_rate": 2.0621991209097888e-05,
+      "loss": 0.358,
+      "step": 10400
+    },
+    {
+      "epoch": 34.88,
+      "learning_rate": 2.0517337612502618e-05,
+      "loss": 0.3425,
+      "step": 10500
+    },
+    {
+      "epoch": 35.0,
+      "eval_loss": 0.3589760661125183,
+      "eval_runtime": 109.5451,
+      "eval_samples_per_second": 13.693,
+      "eval_wer": 0.3215408386168682,
+      "step": 10535
+    },
+    {
+      "epoch": 35.22,
+      "learning_rate": 2.0412684015907345e-05,
+      "loss": 0.3439,
+      "step": 10600
+    },
+    {
+      "epoch": 35.55,
+      "learning_rate": 2.0308030419312078e-05,
+      "loss": 0.3396,
+      "step": 10700
+    },
+    {
+      "epoch": 35.88,
+      "learning_rate": 2.0203376822716808e-05,
+      "loss": 0.3434,
+      "step": 10800
+    },
+    {
+      "epoch": 36.0,
+      "eval_loss": 0.3650296926498413,
+      "eval_runtime": 110.2908,
+      "eval_samples_per_second": 13.6,
+      "eval_wer": 0.3190899888113378,
+      "step": 10836
+    },
+    {
+      "epoch": 36.21,
+      "learning_rate": 2.0098723226121538e-05,
+      "loss": 0.339,
+      "step": 10900
+    },
+    {
+      "epoch": 36.54,
+      "learning_rate": 1.9994069629526268e-05,
+      "loss": 0.3369,
+      "step": 11000
+    },
+    {
+      "epoch": 36.88,
+      "learning_rate": 1.9889416032931e-05,
+      "loss": 0.3381,
+      "step": 11100
+    },
+    {
+      "epoch": 37.0,
+      "eval_loss": 0.36705273389816284,
+      "eval_runtime": 109.552,
+      "eval_samples_per_second": 13.692,
+      "eval_wer": 0.31930310618573177,
+      "step": 11137
+    },
+    {
+      "epoch": 37.21,
+      "learning_rate": 1.9784762436335728e-05,
+      "loss": 0.3315,
+      "step": 11200
+    },
+    {
+      "epoch": 37.54,
+      "learning_rate": 1.9680108839740458e-05,
+      "loss": 0.333,
+      "step": 11300
+    },
+    {
+      "epoch": 37.87,
+      "learning_rate": 1.957545524314519e-05,
+      "loss": 0.3269,
+      "step": 11400
+    },
+    {
+      "epoch": 38.0,
+      "eval_loss": 0.34720638394355774,
+      "eval_runtime": 109.9976,
+      "eval_samples_per_second": 13.637,
+      "eval_wer": 0.3030529063881933,
+      "step": 11438
+    },
+    {
+      "epoch": 38.21,
+      "learning_rate": 1.947080164654992e-05,
+      "loss": 0.3335,
+      "step": 11500
+    },
+    {
+      "epoch": 38.54,
+      "learning_rate": 1.936614804995465e-05,
+      "loss": 0.3293,
+      "step": 11600
+    },
+    {
+      "epoch": 38.87,
+      "learning_rate": 1.926149445335938e-05,
+      "loss": 0.3241,
+      "step": 11700
+    },
+    {
+      "epoch": 39.0,
+      "eval_loss": 0.360032320022583,
+      "eval_runtime": 110.7657,
+      "eval_samples_per_second": 13.542,
+      "eval_wer": 0.3122702328307315,
+      "step": 11739
+    },
+    {
+      "epoch": 39.2,
+      "learning_rate": 1.9156840856764114e-05,
+      "loss": 0.328,
+      "step": 11800
+    },
+    {
+      "epoch": 39.53,
+      "learning_rate": 1.905218726016884e-05,
+      "loss": 0.3253,
+      "step": 11900
+    },
+    {
+      "epoch": 39.87,
+      "learning_rate": 1.894753366357357e-05,
+      "loss": 0.3249,
+      "step": 12000
+    },
+    {
+      "epoch": 40.0,
+      "eval_loss": 0.37267357110977173,
+      "eval_runtime": 110.521,
+      "eval_samples_per_second": 13.572,
+      "eval_wer": 0.3121636741435346,
+      "step": 12040
+    },
+    {
+      "epoch": 40.2,
+      "learning_rate": 1.8842880066978304e-05,
+      "loss": 0.3181,
+      "step": 12100
+    },
+    {
+      "epoch": 40.53,
+      "learning_rate": 1.873822647038303e-05,
+      "loss": 0.313,
+      "step": 12200
+    },
+    {
+      "epoch": 40.86,
+      "learning_rate": 1.8633572873787764e-05,
+      "loss": 0.3161,
+      "step": 12300
+    },
+    {
+      "epoch": 41.0,
+      "eval_loss": 0.35508450865745544,
+      "eval_runtime": 110.5241,
+      "eval_samples_per_second": 13.572,
+      "eval_wer": 0.31093824924076935,
+      "step": 12341
+    },
+    {
+      "epoch": 41.2,
+      "learning_rate": 1.8528919277192494e-05,
+      "loss": 0.3106,
+      "step": 12400
+    },
+    {
+      "epoch": 41.53,
+      "learning_rate": 1.842426568059722e-05,
+      "loss": 0.3111,
+      "step": 12500
+    },
+    {
+      "epoch": 41.86,
+      "learning_rate": 1.8319612084001954e-05,
+      "loss": 0.3084,
+      "step": 12600
+    },
+    {
+      "epoch": 42.0,
+      "eval_loss": 0.3433784246444702,
+      "eval_runtime": 110.4783,
+      "eval_samples_per_second": 13.577,
+      "eval_wer": 0.29388885928925357,
+      "step": 12642
+    },
+    {
+      "epoch": 42.19,
+      "learning_rate": 1.8214958487406684e-05,
+      "loss": 0.3099,
+      "step": 12700
+    },
+    {
+      "epoch": 42.52,
+      "learning_rate": 1.8110304890811414e-05,
+      "loss": 0.3091,
+      "step": 12800
+    },
+    {
+      "epoch": 42.86,
+      "learning_rate": 1.8005651294216144e-05,
+      "loss": 0.3073,
+      "step": 12900
+    },
+    {
+      "epoch": 43.0,
+      "eval_loss": 0.35515299439430237,
+      "eval_runtime": 108.9447,
+      "eval_samples_per_second": 13.768,
+      "eval_wer": 0.29681922318717036,
+      "step": 12943
+    },
+    {
+      "epoch": 43.19,
+      "learning_rate": 1.7900997697620877e-05,
+      "loss": 0.3048,
+      "step": 13000
+    },
+    {
+      "epoch": 43.52,
+      "learning_rate": 1.7796344101025604e-05,
+      "loss": 0.3043,
+      "step": 13100
+    },
+    {
+      "epoch": 43.85,
+      "learning_rate": 1.7691690504430334e-05,
+      "loss": 0.2989,
+      "step": 13200
+    },
+    {
+      "epoch": 44.0,
+      "eval_loss": 0.3512587547302246,
+      "eval_runtime": 108.2231,
+      "eval_samples_per_second": 13.86,
+      "eval_wer": 0.2980446480899355,
+      "step": 13244
+    },
+    {
+      "epoch": 44.19,
+      "learning_rate": 1.7587036907835067e-05,
+      "loss": 0.3008,
+      "step": 13300
+    },
+    {
+      "epoch": 44.52,
+      "learning_rate": 1.7482383311239797e-05,
+      "loss": 0.2991,
+      "step": 13400
+    },
+    {
+      "epoch": 44.85,
+      "learning_rate": 1.7377729714644527e-05,
+      "loss": 0.3037,
+      "step": 13500
+    },
+    {
+      "epoch": 45.0,
+      "eval_loss": 0.34435179829597473,
+      "eval_runtime": 108.9568,
+      "eval_samples_per_second": 13.767,
+      "eval_wer": 0.29511428419201874,
+      "step": 13545
+    },
+    {
+      "epoch": 45.18,
+      "learning_rate": 1.7273076118049257e-05,
+      "loss": 0.3016,
+      "step": 13600
+    },
+    {
+      "epoch": 45.51,
+      "learning_rate": 1.716842252145399e-05,
+      "loss": 0.2948,
+      "step": 13700
+    },
+    {
+      "epoch": 45.85,
+      "learning_rate": 1.7063768924858717e-05,
+      "loss": 0.2927,
+      "step": 13800
+    },
+    {
+      "epoch": 46.0,
+      "eval_loss": 0.33969631791114807,
+      "eval_runtime": 107.6041,
+      "eval_samples_per_second": 13.94,
+      "eval_wer": 0.2929298311044808,
+      "step": 13846
+    },
+    {
+      "epoch": 46.18,
+      "learning_rate": 1.6959115328263447e-05,
+      "loss": 0.2942,
+      "step": 13900
+    },
+    {
+      "epoch": 46.51,
+      "learning_rate": 1.685446173166818e-05,
+      "loss": 0.2932,
+      "step": 14000
+    },
+    {
+      "epoch": 46.84,
+      "learning_rate": 1.6749808135072907e-05,
+      "loss": 0.2856,
+      "step": 14100
+    },
+    {
+      "epoch": 47.0,
+      "eval_loss": 0.34951111674308777,
+      "eval_runtime": 109.88,
+      "eval_samples_per_second": 13.651,
+      "eval_wer": 0.2958069156587991,
+      "step": 14147
+    },
+    {
+      "epoch": 47.18,
+      "learning_rate": 1.664515453847764e-05,
+      "loss": 0.2923,
+      "step": 14200
+    },
+    {
+      "epoch": 47.51,
+      "learning_rate": 1.654050094188237e-05,
+      "loss": 0.2902,
+      "step": 14300
+    },
+    {
+      "epoch": 47.84,
+      "learning_rate": 1.64358473452871e-05,
+      "loss": 0.298,
+      "step": 14400
+    },
+    {
+      "epoch": 48.0,
+      "eval_loss": 0.32980483770370483,
+      "eval_runtime": 110.351,
+      "eval_samples_per_second": 13.593,
+      "eval_wer": 0.28685598593425327,
+      "step": 14448
+    },
+    {
+      "epoch": 48.17,
+      "learning_rate": 1.633119374869183e-05,
+      "loss": 0.2849,
+      "step": 14500
+    },
+    {
+      "epoch": 48.5,
+      "learning_rate": 1.622654015209656e-05,
+      "loss": 0.2858,
+      "step": 14600
+    },
+    {
+      "epoch": 48.84,
+      "learning_rate": 1.612188655550129e-05,
+      "loss": 0.2808,
+      "step": 14700
+    },
+    {
+      "epoch": 49.0,
+      "eval_loss": 0.33403417468070984,
+      "eval_runtime": 109.6552,
+      "eval_samples_per_second": 13.679,
+      "eval_wer": 0.2835526666311471,
+      "step": 14749
+    },
+    {
+      "epoch": 49.17,
+      "learning_rate": 1.601723295890602e-05,
+      "loss": 0.285,
+      "step": 14800
+    },
+    {
+      "epoch": 49.5,
+      "learning_rate": 1.5912579362310754e-05,
+      "loss": 0.2846,
+      "step": 14900
+    },
+    {
+      "epoch": 49.83,
+      "learning_rate": 1.580792576571548e-05,
+      "loss": 0.28,
+      "step": 15000
+    },
+    {
+      "epoch": 50.0,
+      "eval_loss": 0.3472961485385895,
+      "eval_runtime": 109.7218,
+      "eval_samples_per_second": 13.671,
+      "eval_wer": 0.2827534764771698,
+      "step": 15050
+    },
+    {
+      "epoch": 50.17,
+      "learning_rate": 1.5703272169120214e-05,
+      "loss": 0.2812,
+      "step": 15100
+    },
+    {
+      "epoch": 50.5,
+      "learning_rate": 1.5598618572524944e-05,
+      "loss": 0.2797,
+      "step": 15200
+    },
+    {
+      "epoch": 50.83,
+      "learning_rate": 1.5493964975929674e-05,
+      "loss": 0.2749,
+      "step": 15300
+    },
+    {
+      "epoch": 51.0,
+      "eval_loss": 0.3344120681285858,
+      "eval_runtime": 110.2377,
+      "eval_samples_per_second": 13.607,
+      "eval_wer": 0.281794448292397,
+      "step": 15351
+    },
+    {
+      "epoch": 51.16,
+      "learning_rate": 1.5389311379334404e-05,
+      "loss": 0.2803,
+      "step": 15400
+    },
+    {
+      "epoch": 51.49,
+      "learning_rate": 1.5284657782739134e-05,
+      "loss": 0.2737,
+      "step": 15500
+    },
+    {
+      "epoch": 51.83,
+      "learning_rate": 1.5180004186143865e-05,
+      "loss": 0.2771,
+      "step": 15600
+    },
+    {
+      "epoch": 52.0,
+      "eval_loss": 0.33861950039863586,
+      "eval_runtime": 109.5924,
+      "eval_samples_per_second": 13.687,
+      "eval_wer": 0.28243380041557886,
+      "step": 15652
+    },
+    {
+      "epoch": 52.16,
+      "learning_rate": 1.5075350589548594e-05,
+      "loss": 0.2714,
+      "step": 15700
+    },
+    {
+      "epoch": 52.49,
+      "learning_rate": 1.4970696992953325e-05,
+      "loss": 0.2715,
+      "step": 15800
+    },
+    {
+      "epoch": 52.82,
+      "learning_rate": 1.4866043396358055e-05,
+      "loss": 0.2712,
+      "step": 15900
+    },
+    {
+      "epoch": 53.0,
+      "eval_loss": 0.35900208353996277,
+      "eval_runtime": 109.166,
+      "eval_samples_per_second": 13.741,
+      "eval_wer": 0.2847780915339123,
+      "step": 15953
+    },
+    {
+      "epoch": 53.16,
+      "learning_rate": 1.4761389799762787e-05,
+      "loss": 0.2695,
+      "step": 16000
+    },
+    {
+      "epoch": 53.49,
+      "learning_rate": 1.4656736203167517e-05,
+      "loss": 0.2725,
+      "step": 16100
+    },
+    {
+      "epoch": 53.82,
+      "learning_rate": 1.4552082606572245e-05,
+      "loss": 0.2691,
+      "step": 16200
+    },
+    {
+      "epoch": 54.0,
+      "eval_loss": 0.34280407428741455,
+      "eval_runtime": 110.4941,
+      "eval_samples_per_second": 13.575,
+      "eval_wer": 0.2774788214609196,
+      "step": 16254
+    },
+    {
+      "epoch": 54.15,
+      "learning_rate": 1.4447429009976977e-05,
+      "loss": 0.2675,
+      "step": 16300
+    },
+    {
+      "epoch": 54.48,
+      "learning_rate": 1.4342775413381707e-05,
+      "loss": 0.2703,
+      "step": 16400
+    },
+    {
+      "epoch": 54.82,
+      "learning_rate": 1.4238121816786437e-05,
+      "loss": 0.2639,
+      "step": 16500
+    },
+    {
+      "epoch": 55.0,
+      "eval_loss": 0.34979143738746643,
+      "eval_runtime": 110.0168,
+      "eval_samples_per_second": 13.634,
+      "eval_wer": 0.2896265118013746,
+      "step": 16555
+    },
+    {
+      "epoch": 55.15,
+      "learning_rate": 1.4133468220191168e-05,
+      "loss": 0.2698,
+      "step": 16600
+    },
+    {
+      "epoch": 55.48,
+      "learning_rate": 1.4029861159561852e-05,
+      "loss": 0.262,
+      "step": 16700
+    },
+    {
+      "epoch": 55.81,
+      "learning_rate": 1.3925207562966582e-05,
+      "loss": 0.2626,
+      "step": 16800
+    },
+    {
+      "epoch": 56.0,
+      "eval_loss": 0.3468811810016632,
+      "eval_runtime": 110.3507,
+      "eval_samples_per_second": 13.593,
+      "eval_wer": 0.27572060312216956,
+      "step": 16856
+    },
+    {
+      "epoch": 56.15,
+      "learning_rate": 1.3820553966371312e-05,
+      "loss": 0.2641,
+      "step": 16900
+    },
+    {
+      "epoch": 56.48,
+      "learning_rate": 1.3715900369776042e-05,
+      "loss": 0.2608,
+      "step": 17000
+    },
+    {
+      "epoch": 56.81,
+      "learning_rate": 1.3611246773180772e-05,
+      "loss": 0.2656,
+      "step": 17100
+    },
+    {
+      "epoch": 57.0,
+      "eval_loss": 0.33657267689704895,
+      "eval_runtime": 110.9329,
+      "eval_samples_per_second": 13.522,
+      "eval_wer": 0.2681549363311844,
+      "step": 17157
+    },
+    {
+      "epoch": 57.14,
+      "learning_rate": 1.3506593176585502e-05,
+      "loss": 0.262,
+      "step": 17200
+    },
+    {
+      "epoch": 57.47,
+      "learning_rate": 1.3401939579990233e-05,
+      "loss": 0.2596,
+      "step": 17300
+    },
+    {
+      "epoch": 57.81,
+      "learning_rate": 1.3297285983394963e-05,
+      "loss": 0.2557,
+      "step": 17400
+    },
+    {
+      "epoch": 58.0,
+      "eval_loss": 0.3309778571128845,
+      "eval_runtime": 108.9984,
+      "eval_samples_per_second": 13.762,
+      "eval_wer": 0.27758538014811657,
+      "step": 17458
+    },
+    {
+      "epoch": 58.0,
+      "step": 17458,
+      "total_flos": 0,
+      "train_runtime": 265934.449,
+      "train_samples_per_second": 0.113
+    }
+  ],
+  "max_steps": 30100,
+  "num_train_epochs": 100,
+  "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:ba0f1ea343b8845166cb2af28518753f12cd32b454586ef01b2c17630e8c5c17
+size 2543

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}