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 in Russian using the Common Voice 7.0, MAILABS plus data augmentation
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    metrics:
+       - name: Test Common Voice 7.0 WER
+         type: wer
+         value: 19.46
+---
+
+# Wav2vec2 Large 100k Voxpopuli fine-tuned in Russian using the Common Voice 7.0, MAILABS plus data augmentation
+
+[Wav2vec2 Large 100k Voxpopuli](https://huggingface.co/facebook/wav2vec2-large-100k-voxpopuli) Wav2vec2 Large 100k Voxpopuli fine-tuned in Russian using the Common Voice 7.0, M-AILABS plus 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-Common_Voice_plus_TTS-Dataset_plus_Data_Augmentation-russian")
+
+model = Wav2Vec2ForCTC.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-Common_Voice_plus_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 +1,11 @@
+{
+    "epoch": 73.0,
+    "eval_loss": 0.3584013879299164,
+    "eval_runtime": 316.7442,
+    "eval_samples": 8422,
+    "eval_samples_per_second": 26.589,
+    "eval_wer": 0.3019012195279999,
+    "train_runtime": 465541.7438,
+    "train_samples": 81775,
+    "train_samples_per_second": 0.091
+}

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 +1,192 @@
+{
+    "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":"../../datasets/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":"../../datasets/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":"../../datasets/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": "../../datasets/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/GT+GEN-dxg1/RU/100-epoch/",
+    // CACHE
+    "dataset_cache": "../datasets/",
+
+    // DATASETS
+    "datasets":{ 
+        
+        "files_path": "/workspace/edresson/datasets/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/edresson/datasets/Common_Voice/cv-corpus-7.0-2021-07-21/ru/train_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }, 
+            {                
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/edresson/datasets/M-AILABS/ru_RU/train_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            },
+            {
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/edresson/datasets/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/edresson/datasets/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/edresson/datasets/M-AILABS/ru_RU/train_converted.csv"], // csv files
+                "text_column": "text",
+                "path_column": "file_path"
+            }
+            ]
+        ,
+        "devel":
+            [ 
+                {
+                    "name": "csv",
+                    "path": "csv",
+                    "data_files": ["/workspace/edresson/datasets/Common_Voice/cv-corpus-7.0-2021-07-21/ru/dev_converted.csv"], // csv files
+                    "text_column": "text",
+                    "path_column": "file_path"
+                }
+        
+            ] 
+            ,
+        "test":
+            {
+                "name": "csv",
+                "path": "csv",
+                "data_files": ["/workspace/edresson/datasets/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 +1,8 @@
+{
+    "epoch": 73.0,
+    "eval_loss": 0.3584013879299164,
+    "eval_runtime": 316.7442,
+    "eval_samples": 8422,
+    "eval_samples_per_second": 26.589,
+    "eval_wer": 0.3019012195279999
+}

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:d0cac17b90fe449c3bd7d02c54800cebdc62810f425a9eaafb04b23b6e2df058
+size 1262088988

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 +1,6 @@
+{
+    "epoch": 73.0,
+    "train_runtime": 465541.7438,
+    "train_samples": 81775,
+    "train_samples_per_second": 0.091
+}

trainer_state.json

@@ -0,0 +1,2473 @@
+{
+  "best_metric": 0.34823477268218994,
+  "best_model_checkpoint": "../checkpoints/Wav2Vec-voxpopuli/one-speaker/Final-paper/GT+GEN-dxg1/RU/100-epoch/checkpoint-26775",
+  "epoch": 72.99784777929955,
+  "global_step": 31025,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.4822134387351778e-08,
+      "loss": 12.3693,
+      "step": 1
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.4525691699604743e-06,
+      "loss": 16.0709,
+      "step": 100
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.9347826086956523e-06,
+      "loss": 15.3276,
+      "step": 200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 4.416996047430831e-06,
+      "loss": 12.0358,
+      "step": 300
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.8843873517786565e-06,
+      "loss": 9.1409,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 6.154871940612793,
+      "eval_runtime": 318.3436,
+      "eval_samples_per_second": 26.456,
+      "eval_wer": 1.0,
+      "step": 425
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 7.366600790513834e-06,
+      "loss": 7.4893,
+      "step": 500
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 8.848814229249012e-06,
+      "loss": 6.0766,
+      "step": 600
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.033102766798419e-05,
+      "loss": 5.2545,
+      "step": 700
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1.1813241106719368e-05,
+      "loss": 4.5575,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 3.599003553390503,
+      "eval_runtime": 317.6472,
+      "eval_samples_per_second": 26.514,
+      "eval_wer": 1.0,
+      "step": 850
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1.3295454545454546e-05,
+      "loss": 4.0577,
+      "step": 900
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1.4777667984189723e-05,
+      "loss": 3.5943,
+      "step": 1000
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1.6259881422924904e-05,
+      "loss": 3.3759,
+      "step": 1100
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1.774209486166008e-05,
+      "loss": 3.2785,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 3.146242380142212,
+      "eval_runtime": 318.9247,
+      "eval_samples_per_second": 26.407,
+      "eval_wer": 1.0,
+      "step": 1275
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 1.9224308300395258e-05,
+      "loss": 3.2268,
+      "step": 1300
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 2.0706521739130433e-05,
+      "loss": 3.1416,
+      "step": 1400
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 2.2188735177865615e-05,
+      "loss": 3.1018,
+      "step": 1500
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 2.3670948616600794e-05,
+      "loss": 3.0594,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 2.515316205533597e-05,
+      "loss": 3.0063,
+      "step": 1700
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 2.9775705337524414,
+      "eval_runtime": 316.7007,
+      "eval_samples_per_second": 26.593,
+      "eval_wer": 1.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 2.6635375494071148e-05,
+      "loss": 2.8872,
+      "step": 1800
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 2.8117588932806323e-05,
+      "loss": 2.5959,
+      "step": 1900
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 2.9599802371541505e-05,
+      "loss": 2.1892,
+      "step": 2000
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 2.9945893863029945e-05,
+      "loss": 1.9088,
+      "step": 2100
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 1.6493474245071411,
+      "eval_runtime": 318.4422,
+      "eval_samples_per_second": 26.448,
+      "eval_wer": 0.9566750042119724,
+      "step": 2125
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 2.987177586718055e-05,
+      "loss": 1.6933,
+      "step": 2200
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 2.979765787133116e-05,
+      "loss": 1.5286,
+      "step": 2300
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 2.9723539875481766e-05,
+      "loss": 1.4406,
+      "step": 2400
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 2.9649421879632374e-05,
+      "loss": 1.3302,
+      "step": 2500
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 1.2176637649536133,
+      "eval_runtime": 318.9899,
+      "eval_samples_per_second": 26.402,
+      "eval_wer": 0.8561190238592035,
+      "step": 2550
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 2.9575303883782986e-05,
+      "loss": 1.251,
+      "step": 2600
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 2.950118588793359e-05,
+      "loss": 1.162,
+      "step": 2700
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 2.94270678920842e-05,
+      "loss": 1.1033,
+      "step": 2800
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 2.9352949896234806e-05,
+      "loss": 1.0447,
+      "step": 2900
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 0.9863032102584839,
+      "eval_runtime": 317.5697,
+      "eval_samples_per_second": 26.52,
+      "eval_wer": 0.7464522232734153,
+      "step": 2975
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 2.9278831900385414e-05,
+      "loss": 1.0038,
+      "step": 3000
+    },
+    {
+      "epoch": 7.29,
+      "learning_rate": 2.9204713904536022e-05,
+      "loss": 0.963,
+      "step": 3100
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 2.913059590868663e-05,
+      "loss": 0.9284,
+      "step": 3200
+    },
+    {
+      "epoch": 7.76,
+      "learning_rate": 2.9056477912837235e-05,
+      "loss": 0.8925,
+      "step": 3300
+    },
+    {
+      "epoch": 8.0,
+      "learning_rate": 2.8982359916987847e-05,
+      "loss": 0.8712,
+      "step": 3400
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 0.8418087363243103,
+      "eval_runtime": 318.12,
+      "eval_samples_per_second": 26.474,
+      "eval_wer": 0.653970269955029,
+      "step": 3400
+    },
+    {
+      "epoch": 8.23,
+      "learning_rate": 2.890824192113845e-05,
+      "loss": 0.8478,
+      "step": 3500
+    },
+    {
+      "epoch": 8.47,
+      "learning_rate": 2.883412392528906e-05,
+      "loss": 0.8175,
+      "step": 3600
+    },
+    {
+      "epoch": 8.7,
+      "learning_rate": 2.876000592943967e-05,
+      "loss": 0.8035,
+      "step": 3700
+    },
+    {
+      "epoch": 8.94,
+      "learning_rate": 2.8685887933590275e-05,
+      "loss": 0.7815,
+      "step": 3800
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 0.7524622678756714,
+      "eval_runtime": 318.9732,
+      "eval_samples_per_second": 26.403,
+      "eval_wer": 0.5982426355283109,
+      "step": 3825
+    },
+    {
+      "epoch": 9.18,
+      "learning_rate": 2.8611769937740884e-05,
+      "loss": 0.7671,
+      "step": 3900
+    },
+    {
+      "epoch": 9.41,
+      "learning_rate": 2.853765194189149e-05,
+      "loss": 0.7474,
+      "step": 4000
+    },
+    {
+      "epoch": 9.65,
+      "learning_rate": 2.84635339460421e-05,
+      "loss": 0.7477,
+      "step": 4100
+    },
+    {
+      "epoch": 9.88,
+      "learning_rate": 2.8389415950192708e-05,
+      "loss": 0.7128,
+      "step": 4200
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 0.6934680938720703,
+      "eval_runtime": 317.9061,
+      "eval_samples_per_second": 26.492,
+      "eval_wer": 0.5617993545962339,
+      "step": 4250
+    },
+    {
+      "epoch": 10.12,
+      "learning_rate": 2.8315297954343316e-05,
+      "loss": 0.7134,
+      "step": 4300
+    },
+    {
+      "epoch": 10.35,
+      "learning_rate": 2.8241179958493924e-05,
+      "loss": 0.7025,
+      "step": 4400
+    },
+    {
+      "epoch": 10.59,
+      "learning_rate": 2.8167061962644532e-05,
+      "loss": 0.6806,
+      "step": 4500
+    },
+    {
+      "epoch": 10.82,
+      "learning_rate": 2.8092943966795137e-05,
+      "loss": 0.6713,
+      "step": 4600
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 0.6441113948822021,
+      "eval_runtime": 330.7931,
+      "eval_samples_per_second": 25.46,
+      "eval_wer": 0.5259781495833388,
+      "step": 4675
+    },
+    {
+      "epoch": 11.06,
+      "learning_rate": 2.8018825970945745e-05,
+      "loss": 0.6707,
+      "step": 4700
+    },
+    {
+      "epoch": 11.29,
+      "learning_rate": 2.7944707975096356e-05,
+      "loss": 0.6552,
+      "step": 4800
+    },
+    {
+      "epoch": 11.53,
+      "learning_rate": 2.787058997924696e-05,
+      "loss": 0.6435,
+      "step": 4900
+    },
+    {
+      "epoch": 11.76,
+      "learning_rate": 2.779647198339757e-05,
+      "loss": 0.632,
+      "step": 5000
+    },
+    {
+      "epoch": 12.0,
+      "learning_rate": 2.7722353987548177e-05,
+      "loss": 0.632,
+      "step": 5100
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 0.6222620010375977,
+      "eval_runtime": 318.5241,
+      "eval_samples_per_second": 26.441,
+      "eval_wer": 0.5066678762587318,
+      "step": 5100
+    },
+    {
+      "epoch": 12.23,
+      "learning_rate": 2.7648235991698785e-05,
+      "loss": 0.6169,
+      "step": 5200
+    },
+    {
+      "epoch": 12.47,
+      "learning_rate": 2.7574117995849393e-05,
+      "loss": 0.6068,
+      "step": 5300
+    },
+    {
+      "epoch": 12.7,
+      "learning_rate": 2.75e-05,
+      "loss": 0.6036,
+      "step": 5400
+    },
+    {
+      "epoch": 12.94,
+      "learning_rate": 2.742588200415061e-05,
+      "loss": 0.5917,
+      "step": 5500
+    },
+    {
+      "epoch": 13.0,
+      "eval_loss": 0.5906757712364197,
+      "eval_runtime": 319.1883,
+      "eval_samples_per_second": 26.386,
+      "eval_wer": 0.4835214680991693,
+      "step": 5525
+    },
+    {
+      "epoch": 13.18,
+      "learning_rate": 2.7351764008301217e-05,
+      "loss": 0.5905,
+      "step": 5600
+    },
+    {
+      "epoch": 13.41,
+      "learning_rate": 2.7277646012451822e-05,
+      "loss": 0.5818,
+      "step": 5700
+    },
+    {
+      "epoch": 13.65,
+      "learning_rate": 2.720352801660243e-05,
+      "loss": 0.5737,
+      "step": 5800
+    },
+    {
+      "epoch": 13.88,
+      "learning_rate": 2.712941002075304e-05,
+      "loss": 0.5607,
+      "step": 5900
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 0.5859212279319763,
+      "eval_runtime": 319.205,
+      "eval_samples_per_second": 26.384,
+      "eval_wer": 0.47469576599577507,
+      "step": 5950
+    },
+    {
+      "epoch": 14.12,
+      "learning_rate": 2.7055292024903646e-05,
+      "loss": 0.5661,
+      "step": 6000
+    },
+    {
+      "epoch": 14.35,
+      "learning_rate": 2.6981174029054254e-05,
+      "loss": 0.5503,
+      "step": 6100
+    },
+    {
+      "epoch": 14.59,
+      "learning_rate": 2.6907056033204865e-05,
+      "loss": 0.5517,
+      "step": 6200
+    },
+    {
+      "epoch": 14.82,
+      "learning_rate": 2.683293803735547e-05,
+      "loss": 0.544,
+      "step": 6300
+    },
+    {
+      "epoch": 15.0,
+      "eval_loss": 0.5427600741386414,
+      "eval_runtime": 317.4374,
+      "eval_samples_per_second": 26.531,
+      "eval_wer": 0.45460789777219063,
+      "step": 6375
+    },
+    {
+      "epoch": 15.06,
+      "learning_rate": 2.6758820041506078e-05,
+      "loss": 0.5377,
+      "step": 6400
+    },
+    {
+      "epoch": 15.29,
+      "learning_rate": 2.6684702045656686e-05,
+      "loss": 0.5364,
+      "step": 6500
+    },
+    {
+      "epoch": 15.53,
+      "learning_rate": 2.6610584049807294e-05,
+      "loss": 0.5207,
+      "step": 6600
+    },
+    {
+      "epoch": 15.76,
+      "learning_rate": 2.6536466053957902e-05,
+      "loss": 0.518,
+      "step": 6700
+    },
+    {
+      "epoch": 16.0,
+      "learning_rate": 2.6462348058108507e-05,
+      "loss": 0.5208,
+      "step": 6800
+    },
+    {
+      "epoch": 16.0,
+      "eval_loss": 0.549843966960907,
+      "eval_runtime": 315.364,
+      "eval_samples_per_second": 26.706,
+      "eval_wer": 0.4465857103977398,
+      "step": 6800
+    },
+    {
+      "epoch": 16.23,
+      "learning_rate": 2.638823006225912e-05,
+      "loss": 0.5157,
+      "step": 6900
+    },
+    {
+      "epoch": 16.47,
+      "learning_rate": 2.6314112066409726e-05,
+      "loss": 0.5056,
+      "step": 7000
+    },
+    {
+      "epoch": 16.7,
+      "learning_rate": 2.623999407056033e-05,
+      "loss": 0.5052,
+      "step": 7100
+    },
+    {
+      "epoch": 16.94,
+      "learning_rate": 2.616587607471094e-05,
+      "loss": 0.4979,
+      "step": 7200
+    },
+    {
+      "epoch": 17.0,
+      "eval_loss": 0.5181886553764343,
+      "eval_runtime": 316.5373,
+      "eval_samples_per_second": 26.607,
+      "eval_wer": 0.432731561280958,
+      "step": 7225
+    },
+    {
+      "epoch": 17.18,
+      "learning_rate": 2.609175807886155e-05,
+      "loss": 0.4981,
+      "step": 7300
+    },
+    {
+      "epoch": 17.41,
+      "learning_rate": 2.6017640083012155e-05,
+      "loss": 0.4886,
+      "step": 7400
+    },
+    {
+      "epoch": 17.65,
+      "learning_rate": 2.5943522087162763e-05,
+      "loss": 0.4807,
+      "step": 7500
+    },
+    {
+      "epoch": 17.88,
+      "learning_rate": 2.586940409131337e-05,
+      "loss": 0.4829,
+      "step": 7600
+    },
+    {
+      "epoch": 18.0,
+      "eval_loss": 0.5218132138252258,
+      "eval_runtime": 316.5653,
+      "eval_samples_per_second": 26.604,
+      "eval_wer": 0.42257098793431913,
+      "step": 7650
+    },
+    {
+      "epoch": 18.12,
+      "learning_rate": 2.579528609546398e-05,
+      "loss": 0.4803,
+      "step": 7700
+    },
+    {
+      "epoch": 18.35,
+      "learning_rate": 2.5721168099614588e-05,
+      "loss": 0.4731,
+      "step": 7800
+    },
+    {
+      "epoch": 18.59,
+      "learning_rate": 2.5647050103765192e-05,
+      "loss": 0.4732,
+      "step": 7900
+    },
+    {
+      "epoch": 18.82,
+      "learning_rate": 2.5572932107915804e-05,
+      "loss": 0.4667,
+      "step": 8000
+    },
+    {
+      "epoch": 19.0,
+      "eval_loss": 0.49254098534584045,
+      "eval_runtime": 314.8242,
+      "eval_samples_per_second": 26.751,
+      "eval_wer": 0.4067339718251448,
+      "step": 8075
+    },
+    {
+      "epoch": 19.06,
+      "learning_rate": 2.5498814112066412e-05,
+      "loss": 0.47,
+      "step": 8100
+    },
+    {
+      "epoch": 19.29,
+      "learning_rate": 2.5424696116217016e-05,
+      "loss": 0.4549,
+      "step": 8200
+    },
+    {
+      "epoch": 19.53,
+      "learning_rate": 2.5350578120367624e-05,
+      "loss": 0.4486,
+      "step": 8300
+    },
+    {
+      "epoch": 19.76,
+      "learning_rate": 2.5276460124518236e-05,
+      "loss": 0.4516,
+      "step": 8400
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 2.520234212866884e-05,
+      "loss": 0.4517,
+      "step": 8500
+    },
+    {
+      "epoch": 20.0,
+      "eval_loss": 0.4848106801509857,
+      "eval_runtime": 317.0239,
+      "eval_samples_per_second": 26.566,
+      "eval_wer": 0.398400746491103,
+      "step": 8500
+    },
+    {
+      "epoch": 20.23,
+      "learning_rate": 2.512822413281945e-05,
+      "loss": 0.4469,
+      "step": 8600
+    },
+    {
+      "epoch": 20.47,
+      "learning_rate": 2.505410613697006e-05,
+      "loss": 0.441,
+      "step": 8700
+    },
+    {
+      "epoch": 20.7,
+      "learning_rate": 2.4979988141120665e-05,
+      "loss": 0.4379,
+      "step": 8800
+    },
+    {
+      "epoch": 20.94,
+      "learning_rate": 2.4905870145271273e-05,
+      "loss": 0.4372,
+      "step": 8900
+    },
+    {
+      "epoch": 21.0,
+      "eval_loss": 0.48364755511283875,
+      "eval_runtime": 316.0715,
+      "eval_samples_per_second": 26.646,
+      "eval_wer": 0.39547180570495455,
+      "step": 8925
+    },
+    {
+      "epoch": 21.18,
+      "learning_rate": 2.4831752149421877e-05,
+      "loss": 0.4374,
+      "step": 9000
+    },
+    {
+      "epoch": 21.41,
+      "learning_rate": 2.475763415357249e-05,
+      "loss": 0.4326,
+      "step": 9100
+    },
+    {
+      "epoch": 21.65,
+      "learning_rate": 2.4683516157723097e-05,
+      "loss": 0.4247,
+      "step": 9200
+    },
+    {
+      "epoch": 21.88,
+      "learning_rate": 2.46093981618737e-05,
+      "loss": 0.4277,
+      "step": 9300
+    },
+    {
+      "epoch": 22.0,
+      "eval_loss": 0.48058849573135376,
+      "eval_runtime": 319.3273,
+      "eval_samples_per_second": 26.374,
+      "eval_wer": 0.39389069607703375,
+      "step": 9350
+    },
+    {
+      "epoch": 22.12,
+      "learning_rate": 2.453528016602431e-05,
+      "loss": 0.4265,
+      "step": 9400
+    },
+    {
+      "epoch": 22.35,
+      "learning_rate": 2.446116217017492e-05,
+      "loss": 0.4147,
+      "step": 9500
+    },
+    {
+      "epoch": 22.59,
+      "learning_rate": 2.4387044174325526e-05,
+      "loss": 0.4194,
+      "step": 9600
+    },
+    {
+      "epoch": 22.82,
+      "learning_rate": 2.4312926178476134e-05,
+      "loss": 0.412,
+      "step": 9700
+    },
+    {
+      "epoch": 23.0,
+      "eval_loss": 0.44829100370407104,
+      "eval_runtime": 316.8142,
+      "eval_samples_per_second": 26.583,
+      "eval_wer": 0.3794274309560529,
+      "step": 9775
+    },
+    {
+      "epoch": 23.06,
+      "learning_rate": 2.4238808182626745e-05,
+      "loss": 0.4157,
+      "step": 9800
+    },
+    {
+      "epoch": 23.29,
+      "learning_rate": 2.416469018677735e-05,
+      "loss": 0.4051,
+      "step": 9900
+    },
+    {
+      "epoch": 23.53,
+      "learning_rate": 2.4090572190927958e-05,
+      "loss": 0.4046,
+      "step": 10000
+    },
+    {
+      "epoch": 23.76,
+      "learning_rate": 2.4016454195078563e-05,
+      "loss": 0.4003,
+      "step": 10100
+    },
+    {
+      "epoch": 24.0,
+      "learning_rate": 2.3942336199229174e-05,
+      "loss": 0.403,
+      "step": 10200
+    },
+    {
+      "epoch": 24.0,
+      "eval_loss": 0.45721083879470825,
+      "eval_runtime": 318.2524,
+      "eval_samples_per_second": 26.463,
+      "eval_wer": 0.37823511877762084,
+      "step": 10200
+    },
+    {
+      "epoch": 24.23,
+      "learning_rate": 2.3868218203379782e-05,
+      "loss": 0.3971,
+      "step": 10300
+    },
+    {
+      "epoch": 24.47,
+      "learning_rate": 2.3794100207530387e-05,
+      "loss": 0.3982,
+      "step": 10400
+    },
+    {
+      "epoch": 24.7,
+      "learning_rate": 2.3719982211680998e-05,
+      "loss": 0.3959,
+      "step": 10500
+    },
+    {
+      "epoch": 24.94,
+      "learning_rate": 2.3645864215831606e-05,
+      "loss": 0.394,
+      "step": 10600
+    },
+    {
+      "epoch": 25.0,
+      "eval_loss": 0.4389103055000305,
+      "eval_runtime": 315.649,
+      "eval_samples_per_second": 26.682,
+      "eval_wer": 0.37056284910770987,
+      "step": 10625
+    },
+    {
+      "epoch": 25.18,
+      "learning_rate": 2.357174621998221e-05,
+      "loss": 0.3935,
+      "step": 10700
+    },
+    {
+      "epoch": 25.41,
+      "learning_rate": 2.349762822413282e-05,
+      "loss": 0.3863,
+      "step": 10800
+    },
+    {
+      "epoch": 25.65,
+      "learning_rate": 2.342351022828343e-05,
+      "loss": 0.3882,
+      "step": 10900
+    },
+    {
+      "epoch": 25.88,
+      "learning_rate": 2.335013341239253e-05,
+      "loss": 0.3887,
+      "step": 11000
+    },
+    {
+      "epoch": 26.0,
+      "eval_loss": 0.43516942858695984,
+      "eval_runtime": 315.434,
+      "eval_samples_per_second": 26.7,
+      "eval_wer": 0.36394033255141844,
+      "step": 11050
+    },
+    {
+      "epoch": 26.12,
+      "learning_rate": 2.3276015416543136e-05,
+      "loss": 0.3902,
+      "step": 11100
+    },
+    {
+      "epoch": 26.35,
+      "learning_rate": 2.3201897420693744e-05,
+      "loss": 0.3858,
+      "step": 11200
+    },
+    {
+      "epoch": 26.59,
+      "learning_rate": 2.3127779424844355e-05,
+      "loss": 0.3785,
+      "step": 11300
+    },
+    {
+      "epoch": 26.82,
+      "learning_rate": 2.305366142899496e-05,
+      "loss": 0.378,
+      "step": 11400
+    },
+    {
+      "epoch": 27.0,
+      "eval_loss": 0.443709135055542,
+      "eval_runtime": 318.0942,
+      "eval_samples_per_second": 26.476,
+      "eval_wer": 0.3679967859410842,
+      "step": 11475
+    },
+    {
+      "epoch": 27.06,
+      "learning_rate": 2.2979543433145568e-05,
+      "loss": 0.3835,
+      "step": 11500
+    },
+    {
+      "epoch": 27.29,
+      "learning_rate": 2.2905425437296176e-05,
+      "loss": 0.3776,
+      "step": 11600
+    },
+    {
+      "epoch": 27.53,
+      "learning_rate": 2.2831307441446784e-05,
+      "loss": 0.3721,
+      "step": 11700
+    },
+    {
+      "epoch": 27.76,
+      "learning_rate": 2.2757189445597392e-05,
+      "loss": 0.3686,
+      "step": 11800
+    },
+    {
+      "epoch": 28.0,
+      "learning_rate": 2.2683071449748e-05,
+      "loss": 0.3711,
+      "step": 11900
+    },
+    {
+      "epoch": 28.0,
+      "eval_loss": 0.4321639835834503,
+      "eval_runtime": 315.6052,
+      "eval_samples_per_second": 26.685,
+      "eval_wer": 0.3558403856870699,
+      "step": 11900
+    },
+    {
+      "epoch": 28.23,
+      "learning_rate": 2.26096946338571e-05,
+      "loss": 0.3767,
+      "step": 12000
+    },
+    {
+      "epoch": 28.47,
+      "learning_rate": 2.253557663800771e-05,
+      "loss": 0.363,
+      "step": 12100
+    },
+    {
+      "epoch": 28.7,
+      "learning_rate": 2.2461458642158317e-05,
+      "loss": 0.3667,
+      "step": 12200
+    },
+    {
+      "epoch": 28.94,
+      "learning_rate": 2.2387340646308925e-05,
+      "loss": 0.3657,
+      "step": 12300
+    },
+    {
+      "epoch": 29.0,
+      "eval_loss": 0.4269185960292816,
+      "eval_runtime": 316.1021,
+      "eval_samples_per_second": 26.643,
+      "eval_wer": 0.3577714130195306,
+      "step": 12325
+    },
+    {
+      "epoch": 29.18,
+      "learning_rate": 2.2313222650459533e-05,
+      "loss": 0.3624,
+      "step": 12400
+    },
+    {
+      "epoch": 29.41,
+      "learning_rate": 2.223910465461014e-05,
+      "loss": 0.3575,
+      "step": 12500
+    },
+    {
+      "epoch": 29.65,
+      "learning_rate": 2.216498665876075e-05,
+      "loss": 0.349,
+      "step": 12600
+    },
+    {
+      "epoch": 29.88,
+      "learning_rate": 2.2090868662911354e-05,
+      "loss": 0.3548,
+      "step": 12700
+    },
+    {
+      "epoch": 30.0,
+      "eval_loss": 0.4217592477798462,
+      "eval_runtime": 317.2188,
+      "eval_samples_per_second": 26.55,
+      "eval_wer": 0.34779227848265315,
+      "step": 12750
+    },
+    {
+      "epoch": 30.12,
+      "learning_rate": 2.2016750667061965e-05,
+      "loss": 0.3589,
+      "step": 12800
+    },
+    {
+      "epoch": 30.35,
+      "learning_rate": 2.194263267121257e-05,
+      "loss": 0.3522,
+      "step": 12900
+    },
+    {
+      "epoch": 30.59,
+      "learning_rate": 2.1868514675363178e-05,
+      "loss": 0.3443,
+      "step": 13000
+    },
+    {
+      "epoch": 30.82,
+      "learning_rate": 2.1794396679513786e-05,
+      "loss": 0.3464,
+      "step": 13100
+    },
+    {
+      "epoch": 31.0,
+      "eval_loss": 0.42884036898612976,
+      "eval_runtime": 317.8415,
+      "eval_samples_per_second": 26.497,
+      "eval_wer": 0.3482329155920737,
+      "step": 13175
+    },
+    {
+      "epoch": 31.06,
+      "learning_rate": 2.1720278683664394e-05,
+      "loss": 0.3489,
+      "step": 13200
+    },
+    {
+      "epoch": 31.29,
+      "learning_rate": 2.1646160687815002e-05,
+      "loss": 0.3443,
+      "step": 13300
+    },
+    {
+      "epoch": 31.53,
+      "learning_rate": 2.157204269196561e-05,
+      "loss": 0.3457,
+      "step": 13400
+    },
+    {
+      "epoch": 31.76,
+      "learning_rate": 2.1497924696116218e-05,
+      "loss": 0.3514,
+      "step": 13500
+    },
+    {
+      "epoch": 32.0,
+      "learning_rate": 2.1423806700266826e-05,
+      "loss": 0.3447,
+      "step": 13600
+    },
+    {
+      "epoch": 32.0,
+      "eval_loss": 0.41752490401268005,
+      "eval_runtime": 317.3655,
+      "eval_samples_per_second": 26.537,
+      "eval_wer": 0.34579645157527766,
+      "step": 13600
+    },
+    {
+      "epoch": 32.23,
+      "learning_rate": 2.1349688704417434e-05,
+      "loss": 0.3373,
+      "step": 13700
+    },
+    {
+      "epoch": 32.47,
+      "learning_rate": 2.127557070856804e-05,
+      "loss": 0.3403,
+      "step": 13800
+    },
+    {
+      "epoch": 32.7,
+      "learning_rate": 2.120145271271865e-05,
+      "loss": 0.3418,
+      "step": 13900
+    },
+    {
+      "epoch": 32.94,
+      "learning_rate": 2.1127334716869255e-05,
+      "loss": 0.3378,
+      "step": 14000
+    },
+    {
+      "epoch": 33.0,
+      "eval_loss": 0.4071788787841797,
+      "eval_runtime": 317.9264,
+      "eval_samples_per_second": 26.49,
+      "eval_wer": 0.3422843146148961,
+      "step": 14025
+    },
+    {
+      "epoch": 33.18,
+      "learning_rate": 2.1053216721019863e-05,
+      "loss": 0.3358,
+      "step": 14100
+    },
+    {
+      "epoch": 33.41,
+      "learning_rate": 2.0979098725170475e-05,
+      "loss": 0.333,
+      "step": 14200
+    },
+    {
+      "epoch": 33.65,
+      "learning_rate": 2.090498072932108e-05,
+      "loss": 0.3339,
+      "step": 14300
+    },
+    {
+      "epoch": 33.88,
+      "learning_rate": 2.0830862733471687e-05,
+      "loss": 0.3323,
+      "step": 14400
+    },
+    {
+      "epoch": 34.0,
+      "eval_loss": 0.39172977209091187,
+      "eval_runtime": 317.2839,
+      "eval_samples_per_second": 26.544,
+      "eval_wer": 0.3331605344668939,
+      "step": 14450
+    },
+    {
+      "epoch": 34.12,
+      "learning_rate": 2.0756744737622295e-05,
+      "loss": 0.3338,
+      "step": 14500
+    },
+    {
+      "epoch": 34.35,
+      "learning_rate": 2.0682626741772903e-05,
+      "loss": 0.3286,
+      "step": 14600
+    },
+    {
+      "epoch": 34.59,
+      "learning_rate": 2.060850874592351e-05,
+      "loss": 0.3286,
+      "step": 14700
+    },
+    {
+      "epoch": 34.82,
+      "learning_rate": 2.053439075007412e-05,
+      "loss": 0.3278,
+      "step": 14800
+    },
+    {
+      "epoch": 35.0,
+      "eval_loss": 0.40219607949256897,
+      "eval_runtime": 317.0822,
+      "eval_samples_per_second": 26.561,
+      "eval_wer": 0.3327587771024222,
+      "step": 14875
+    },
+    {
+      "epoch": 35.06,
+      "learning_rate": 2.0460272754224724e-05,
+      "loss": 0.3245,
+      "step": 14900
+    },
+    {
+      "epoch": 35.29,
+      "learning_rate": 2.0386154758375336e-05,
+      "loss": 0.3184,
+      "step": 15000
+    },
+    {
+      "epoch": 35.53,
+      "learning_rate": 2.031203676252594e-05,
+      "loss": 0.3313,
+      "step": 15100
+    },
+    {
+      "epoch": 35.76,
+      "learning_rate": 2.023791876667655e-05,
+      "loss": 0.3267,
+      "step": 15200
+    },
+    {
+      "epoch": 36.0,
+      "learning_rate": 2.016380077082716e-05,
+      "loss": 0.3221,
+      "step": 15300
+    },
+    {
+      "epoch": 36.0,
+      "eval_loss": 0.4049491584300995,
+      "eval_runtime": 318.0804,
+      "eval_samples_per_second": 26.478,
+      "eval_wer": 0.33557107865372404,
+      "step": 15300
+    },
+    {
+      "epoch": 36.23,
+      "learning_rate": 2.0089682774977765e-05,
+      "loss": 0.3196,
+      "step": 15400
+    },
+    {
+      "epoch": 36.47,
+      "learning_rate": 2.0015564779128373e-05,
+      "loss": 0.3239,
+      "step": 15500
+    },
+    {
+      "epoch": 36.7,
+      "learning_rate": 1.994144678327898e-05,
+      "loss": 0.3199,
+      "step": 15600
+    },
+    {
+      "epoch": 36.94,
+      "learning_rate": 1.986732878742959e-05,
+      "loss": 0.3171,
+      "step": 15700
+    },
+    {
+      "epoch": 37.0,
+      "eval_loss": 0.3955913484096527,
+      "eval_runtime": 317.9907,
+      "eval_samples_per_second": 26.485,
+      "eval_wer": 0.33121654721945026,
+      "step": 15725
+    },
+    {
+      "epoch": 37.18,
+      "learning_rate": 1.9793210791580197e-05,
+      "loss": 0.3235,
+      "step": 15800
+    },
+    {
+      "epoch": 37.41,
+      "learning_rate": 1.9719092795730805e-05,
+      "loss": 0.3155,
+      "step": 15900
+    },
+    {
+      "epoch": 37.65,
+      "learning_rate": 1.9644974799881413e-05,
+      "loss": 0.3193,
+      "step": 16000
+    },
+    {
+      "epoch": 37.88,
+      "learning_rate": 1.957085680403202e-05,
+      "loss": 0.3128,
+      "step": 16100
+    },
+    {
+      "epoch": 38.0,
+      "eval_loss": 0.3912222683429718,
+      "eval_runtime": 317.3702,
+      "eval_samples_per_second": 26.537,
+      "eval_wer": 0.3300112751260352,
+      "step": 16150
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 1.9496738808182626e-05,
+      "loss": 0.315,
+      "step": 16200
+    },
+    {
+      "epoch": 38.35,
+      "learning_rate": 1.9422620812333234e-05,
+      "loss": 0.3106,
+      "step": 16300
+    },
+    {
+      "epoch": 38.59,
+      "learning_rate": 1.9349243996442334e-05,
+      "loss": 0.3091,
+      "step": 16400
+    },
+    {
+      "epoch": 38.82,
+      "learning_rate": 1.9275867180551438e-05,
+      "loss": 0.3114,
+      "step": 16500
+    },
+    {
+      "epoch": 39.0,
+      "eval_loss": 0.40268105268478394,
+      "eval_runtime": 317.2535,
+      "eval_samples_per_second": 26.547,
+      "eval_wer": 0.3303352730006091,
+      "step": 16575
+    },
+    {
+      "epoch": 39.06,
+      "learning_rate": 1.9201749184702046e-05,
+      "loss": 0.3144,
+      "step": 16600
+    },
+    {
+      "epoch": 39.29,
+      "learning_rate": 1.9127631188852654e-05,
+      "loss": 0.3102,
+      "step": 16700
+    },
+    {
+      "epoch": 39.53,
+      "learning_rate": 1.905351319300326e-05,
+      "loss": 0.3098,
+      "step": 16800
+    },
+    {
+      "epoch": 39.76,
+      "learning_rate": 1.897939519715387e-05,
+      "loss": 0.3071,
+      "step": 16900
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 1.890527720130448e-05,
+      "loss": 0.3061,
+      "step": 17000
+    },
+    {
+      "epoch": 40.0,
+      "eval_loss": 0.39571964740753174,
+      "eval_runtime": 319.4169,
+      "eval_samples_per_second": 26.367,
+      "eval_wer": 0.32423115304363603,
+      "step": 17000
+    },
+    {
+      "epoch": 40.23,
+      "learning_rate": 1.8831159205455083e-05,
+      "loss": 0.2991,
+      "step": 17100
+    },
+    {
+      "epoch": 40.47,
+      "learning_rate": 1.8757041209605695e-05,
+      "loss": 0.3006,
+      "step": 17200
+    },
+    {
+      "epoch": 40.7,
+      "learning_rate": 1.8682923213756303e-05,
+      "loss": 0.3041,
+      "step": 17300
+    },
+    {
+      "epoch": 40.94,
+      "learning_rate": 1.8608805217906907e-05,
+      "loss": 0.306,
+      "step": 17400
+    },
+    {
+      "epoch": 41.0,
+      "eval_loss": 0.3856252431869507,
+      "eval_runtime": 318.6166,
+      "eval_samples_per_second": 26.433,
+      "eval_wer": 0.322105726986431,
+      "step": 17425
+    },
+    {
+      "epoch": 41.18,
+      "learning_rate": 1.8534687222057515e-05,
+      "loss": 0.308,
+      "step": 17500
+    },
+    {
+      "epoch": 41.41,
+      "learning_rate": 1.8460569226208123e-05,
+      "loss": 0.298,
+      "step": 17600
+    },
+    {
+      "epoch": 41.65,
+      "learning_rate": 1.838645123035873e-05,
+      "loss": 0.3004,
+      "step": 17700
+    },
+    {
+      "epoch": 41.88,
+      "learning_rate": 1.831233323450934e-05,
+      "loss": 0.299,
+      "step": 17800
+    },
+    {
+      "epoch": 42.0,
+      "eval_loss": 0.384583979845047,
+      "eval_runtime": 316.4886,
+      "eval_samples_per_second": 26.611,
+      "eval_wer": 0.3222223662212776,
+      "step": 17850
+    },
+    {
+      "epoch": 42.12,
+      "learning_rate": 1.8238215238659944e-05,
+      "loss": 0.2965,
+      "step": 17900
+    },
+    {
+      "epoch": 42.35,
+      "learning_rate": 1.8164097242810556e-05,
+      "loss": 0.3002,
+      "step": 18000
+    },
+    {
+      "epoch": 42.59,
+      "learning_rate": 1.8089979246961164e-05,
+      "loss": 0.2961,
+      "step": 18100
+    },
+    {
+      "epoch": 42.82,
+      "learning_rate": 1.801586125111177e-05,
+      "loss": 0.295,
+      "step": 18200
+    },
+    {
+      "epoch": 43.0,
+      "eval_loss": 0.3782486915588379,
+      "eval_runtime": 316.5726,
+      "eval_samples_per_second": 26.604,
+      "eval_wer": 0.3191767862002825,
+      "step": 18275
+    },
+    {
+      "epoch": 43.06,
+      "learning_rate": 1.794174325526238e-05,
+      "loss": 0.2962,
+      "step": 18300
+    },
+    {
+      "epoch": 43.29,
+      "learning_rate": 1.7867625259412988e-05,
+      "loss": 0.2926,
+      "step": 18400
+    },
+    {
+      "epoch": 43.53,
+      "learning_rate": 1.7793507263563593e-05,
+      "loss": 0.2993,
+      "step": 18500
+    },
+    {
+      "epoch": 43.76,
+      "learning_rate": 1.77193892677142e-05,
+      "loss": 0.2908,
+      "step": 18600
+    },
+    {
+      "epoch": 44.0,
+      "learning_rate": 1.764527127186481e-05,
+      "loss": 0.2919,
+      "step": 18700
+    },
+    {
+      "epoch": 44.0,
+      "eval_loss": 0.3738417625427246,
+      "eval_runtime": 316.4549,
+      "eval_samples_per_second": 26.614,
+      "eval_wer": 0.3151980923005145,
+      "step": 18700
+    },
+    {
+      "epoch": 44.23,
+      "learning_rate": 1.7571153276015417e-05,
+      "loss": 0.2927,
+      "step": 18800
+    },
+    {
+      "epoch": 44.47,
+      "learning_rate": 1.7497035280166025e-05,
+      "loss": 0.2913,
+      "step": 18900
+    },
+    {
+      "epoch": 44.7,
+      "learning_rate": 1.7422917284316633e-05,
+      "loss": 0.2906,
+      "step": 19000
+    },
+    {
+      "epoch": 44.94,
+      "learning_rate": 1.734879928846724e-05,
+      "loss": 0.2837,
+      "step": 19100
+    },
+    {
+      "epoch": 45.0,
+      "eval_loss": 0.38119739294052124,
+      "eval_runtime": 317.5894,
+      "eval_samples_per_second": 26.519,
+      "eval_wer": 0.31785487487202085,
+      "step": 19125
+    },
+    {
+      "epoch": 45.18,
+      "learning_rate": 1.727468129261785e-05,
+      "loss": 0.2909,
+      "step": 19200
+    },
+    {
+      "epoch": 45.41,
+      "learning_rate": 1.7200563296768454e-05,
+      "loss": 0.2858,
+      "step": 19300
+    },
+    {
+      "epoch": 45.65,
+      "learning_rate": 1.7126445300919065e-05,
+      "loss": 0.2905,
+      "step": 19400
+    },
+    {
+      "epoch": 45.88,
+      "learning_rate": 1.7052327305069673e-05,
+      "loss": 0.2825,
+      "step": 19500
+    },
+    {
+      "epoch": 46.0,
+      "eval_loss": 0.38637784123420715,
+      "eval_runtime": 317.552,
+      "eval_samples_per_second": 26.522,
+      "eval_wer": 0.31740127784761735,
+      "step": 19550
+    },
+    {
+      "epoch": 46.12,
+      "learning_rate": 1.6978209309220278e-05,
+      "loss": 0.29,
+      "step": 19600
+    },
+    {
+      "epoch": 46.35,
+      "learning_rate": 1.690409131337089e-05,
+      "loss": 0.29,
+      "step": 19700
+    },
+    {
+      "epoch": 46.59,
+      "learning_rate": 1.6829973317521494e-05,
+      "loss": 0.2855,
+      "step": 19800
+    },
+    {
+      "epoch": 46.82,
+      "learning_rate": 1.6755855321672102e-05,
+      "loss": 0.2827,
+      "step": 19900
+    },
+    {
+      "epoch": 47.0,
+      "eval_loss": 0.38330715894699097,
+      "eval_runtime": 317.4978,
+      "eval_samples_per_second": 26.526,
+      "eval_wer": 0.3139798602921165,
+      "step": 19975
+    },
+    {
+      "epoch": 47.06,
+      "learning_rate": 1.668173732582271e-05,
+      "loss": 0.2883,
+      "step": 20000
+    },
+    {
+      "epoch": 47.29,
+      "learning_rate": 1.6607619329973318e-05,
+      "loss": 0.282,
+      "step": 20100
+    },
+    {
+      "epoch": 47.53,
+      "learning_rate": 1.6533501334123926e-05,
+      "loss": 0.2826,
+      "step": 20200
+    },
+    {
+      "epoch": 47.76,
+      "learning_rate": 1.6459383338274534e-05,
+      "loss": 0.2756,
+      "step": 20300
+    },
+    {
+      "epoch": 48.0,
+      "learning_rate": 1.638526534242514e-05,
+      "loss": 0.2849,
+      "step": 20400
+    },
+    {
+      "epoch": 48.0,
+      "eval_loss": 0.3806135356426239,
+      "eval_runtime": 317.6019,
+      "eval_samples_per_second": 26.517,
+      "eval_wer": 0.31536657119529293,
+      "step": 20400
+    },
+    {
+      "epoch": 48.23,
+      "learning_rate": 1.631114734657575e-05,
+      "loss": 0.2844,
+      "step": 20500
+    },
+    {
+      "epoch": 48.47,
+      "learning_rate": 1.623702935072636e-05,
+      "loss": 0.2808,
+      "step": 20600
+    },
+    {
+      "epoch": 48.7,
+      "learning_rate": 1.616365253483546e-05,
+      "loss": 0.2845,
+      "step": 20700
+    },
+    {
+      "epoch": 48.94,
+      "learning_rate": 1.6089534538986064e-05,
+      "loss": 0.2767,
+      "step": 20800
+    },
+    {
+      "epoch": 49.0,
+      "eval_loss": 0.37584495544433594,
+      "eval_runtime": 317.5121,
+      "eval_samples_per_second": 26.525,
+      "eval_wer": 0.3132929847980197,
+      "step": 20825
+    },
+    {
+      "epoch": 49.18,
+      "learning_rate": 1.6015416543136675e-05,
+      "loss": 0.283,
+      "step": 20900
+    },
+    {
+      "epoch": 49.41,
+      "learning_rate": 1.5941298547287283e-05,
+      "loss": 0.2801,
+      "step": 21000
+    },
+    {
+      "epoch": 49.65,
+      "learning_rate": 1.5867180551437888e-05,
+      "loss": 0.2737,
+      "step": 21100
+    },
+    {
+      "epoch": 49.88,
+      "learning_rate": 1.57930625555885e-05,
+      "loss": 0.2732,
+      "step": 21200
+    },
+    {
+      "epoch": 50.0,
+      "eval_loss": 0.3800816833972931,
+      "eval_runtime": 317.0609,
+      "eval_samples_per_second": 26.563,
+      "eval_wer": 0.31414833918689494,
+      "step": 21250
+    },
+    {
+      "epoch": 50.12,
+      "learning_rate": 1.5718944559739107e-05,
+      "loss": 0.2818,
+      "step": 21300
+    },
+    {
+      "epoch": 50.35,
+      "learning_rate": 1.5644826563889712e-05,
+      "loss": 0.2728,
+      "step": 21400
+    },
+    {
+      "epoch": 50.59,
+      "learning_rate": 1.557070856804032e-05,
+      "loss": 0.276,
+      "step": 21500
+    },
+    {
+      "epoch": 50.82,
+      "learning_rate": 1.5496590572190928e-05,
+      "loss": 0.2691,
+      "step": 21600
+    },
+    {
+      "epoch": 51.0,
+      "eval_loss": 0.3639204800128937,
+      "eval_runtime": 317.0756,
+      "eval_samples_per_second": 26.561,
+      "eval_wer": 0.30717590492606367,
+      "step": 21675
+    },
+    {
+      "epoch": 51.06,
+      "learning_rate": 1.5422472576341536e-05,
+      "loss": 0.2741,
+      "step": 21700
+    },
+    {
+      "epoch": 51.29,
+      "learning_rate": 1.5348354580492144e-05,
+      "loss": 0.2722,
+      "step": 21800
+    },
+    {
+      "epoch": 51.53,
+      "learning_rate": 1.527423658464275e-05,
+      "loss": 0.2724,
+      "step": 21900
+    },
+    {
+      "epoch": 51.76,
+      "learning_rate": 1.520011858879336e-05,
+      "loss": 0.2662,
+      "step": 22000
+    },
+    {
+      "epoch": 52.0,
+      "learning_rate": 1.5126000592943967e-05,
+      "loss": 0.2742,
+      "step": 22100
+    },
+    {
+      "epoch": 52.0,
+      "eval_loss": 0.36436760425567627,
+      "eval_runtime": 316.4547,
+      "eval_samples_per_second": 26.614,
+      "eval_wer": 0.30645014968701806,
+      "step": 22100
+    },
+    {
+      "epoch": 52.23,
+      "learning_rate": 1.5051882597094575e-05,
+      "loss": 0.2699,
+      "step": 22200
+    },
+    {
+      "epoch": 52.47,
+      "learning_rate": 1.4977764601245183e-05,
+      "loss": 0.2732,
+      "step": 22300
+    },
+    {
+      "epoch": 52.7,
+      "learning_rate": 1.4903646605395791e-05,
+      "loss": 0.2723,
+      "step": 22400
+    },
+    {
+      "epoch": 52.94,
+      "learning_rate": 1.4829528609546397e-05,
+      "loss": 0.2671,
+      "step": 22500
+    },
+    {
+      "epoch": 53.0,
+      "eval_loss": 0.3704889416694641,
+      "eval_runtime": 317.8613,
+      "eval_samples_per_second": 26.496,
+      "eval_wer": 0.3060743121525123,
+      "step": 22525
+    },
+    {
+      "epoch": 53.18,
+      "learning_rate": 1.4755410613697007e-05,
+      "loss": 0.2713,
+      "step": 22600
+    },
+    {
+      "epoch": 53.41,
+      "learning_rate": 1.4681292617847613e-05,
+      "loss": 0.2658,
+      "step": 22700
+    },
+    {
+      "epoch": 53.65,
+      "learning_rate": 1.4607174621998221e-05,
+      "loss": 0.2707,
+      "step": 22800
+    },
+    {
+      "epoch": 53.88,
+      "learning_rate": 1.4533797806107324e-05,
+      "loss": 0.2618,
+      "step": 22900
+    },
+    {
+      "epoch": 54.0,
+      "eval_loss": 0.3567538857460022,
+      "eval_runtime": 353.5118,
+      "eval_samples_per_second": 23.824,
+      "eval_wer": 0.3049986392089268,
+      "step": 22950
+    },
+    {
+      "epoch": 54.12,
+      "learning_rate": 1.4459679810257932e-05,
+      "loss": 0.2679,
+      "step": 23000
+    },
+    {
+      "epoch": 54.35,
+      "learning_rate": 1.4385561814408538e-05,
+      "loss": 0.2647,
+      "step": 23100
+    },
+    {
+      "epoch": 54.59,
+      "learning_rate": 1.4311443818559146e-05,
+      "loss": 0.2632,
+      "step": 23200
+    },
+    {
+      "epoch": 54.82,
+      "learning_rate": 1.4237325822709754e-05,
+      "loss": 0.265,
+      "step": 23300
+    },
+    {
+      "epoch": 55.0,
+      "eval_loss": 0.3651841878890991,
+      "eval_runtime": 317.4254,
+      "eval_samples_per_second": 26.532,
+      "eval_wer": 0.3071111053511489,
+      "step": 23375
+    },
+    {
+      "epoch": 55.06,
+      "learning_rate": 1.4163207826860362e-05,
+      "loss": 0.2638,
+      "step": 23400
+    },
+    {
+      "epoch": 55.29,
+      "learning_rate": 1.408908983101097e-05,
+      "loss": 0.2648,
+      "step": 23500
+    },
+    {
+      "epoch": 55.53,
+      "learning_rate": 1.4014971835161577e-05,
+      "loss": 0.264,
+      "step": 23600
+    },
+    {
+      "epoch": 55.76,
+      "learning_rate": 1.3940853839312186e-05,
+      "loss": 0.2631,
+      "step": 23700
+    },
+    {
+      "epoch": 56.0,
+      "learning_rate": 1.3866735843462793e-05,
+      "loss": 0.2655,
+      "step": 23800
+    },
+    {
+      "epoch": 56.0,
+      "eval_loss": 0.3658052384853363,
+      "eval_runtime": 316.7135,
+      "eval_samples_per_second": 26.592,
+      "eval_wer": 0.3048431202291313,
+      "step": 23800
+    },
+    {
+      "epoch": 56.23,
+      "learning_rate": 1.3792617847613401e-05,
+      "loss": 0.2635,
+      "step": 23900
+    },
+    {
+      "epoch": 56.47,
+      "learning_rate": 1.3718499851764009e-05,
+      "loss": 0.2658,
+      "step": 24000
+    },
+    {
+      "epoch": 56.7,
+      "learning_rate": 1.3644381855914617e-05,
+      "loss": 0.2683,
+      "step": 24100
+    },
+    {
+      "epoch": 56.94,
+      "learning_rate": 1.3570263860065223e-05,
+      "loss": 0.2588,
+      "step": 24200
+    },
+    {
+      "epoch": 57.0,
+      "eval_loss": 0.3624616861343384,
+      "eval_runtime": 317.8838,
+      "eval_samples_per_second": 26.494,
+      "eval_wer": 0.30356008864581846,
+      "step": 24225
+    },
+    {
+      "epoch": 57.18,
+      "learning_rate": 1.3496145864215831e-05,
+      "loss": 0.2566,
+      "step": 24300
+    },
+    {
+      "epoch": 57.41,
+      "learning_rate": 1.3422027868366441e-05,
+      "loss": 0.2575,
+      "step": 24400
+    },
+    {
+      "epoch": 57.65,
+      "learning_rate": 1.3347909872517047e-05,
+      "loss": 0.2643,
+      "step": 24500
+    },
+    {
+      "epoch": 57.88,
+      "learning_rate": 1.3273791876667656e-05,
+      "loss": 0.2585,
+      "step": 24600
+    },
+    {
+      "epoch": 58.0,
+      "eval_loss": 0.3619252145290375,
+      "eval_runtime": 317.5906,
+      "eval_samples_per_second": 26.518,
+      "eval_wer": 0.3017457005482044,
+      "step": 24650
+    },
+    {
+      "epoch": 58.12,
+      "learning_rate": 1.3199673880818262e-05,
+      "loss": 0.2616,
+      "step": 24700
+    },
+    {
+      "epoch": 58.35,
+      "learning_rate": 1.3125555884968872e-05,
+      "loss": 0.2588,
+      "step": 24800
+    },
+    {
+      "epoch": 58.59,
+      "learning_rate": 1.3051437889119478e-05,
+      "loss": 0.2606,
+      "step": 24900
+    },
+    {
+      "epoch": 58.82,
+      "learning_rate": 1.2977319893270086e-05,
+      "loss": 0.2616,
+      "step": 25000
+    },
+    {
+      "epoch": 59.0,
+      "eval_loss": 0.35793522000312805,
+      "eval_runtime": 316.6098,
+      "eval_samples_per_second": 26.601,
+      "eval_wer": 0.3006829875196019,
+      "step": 25075
+    },
+    {
+      "epoch": 59.06,
+      "learning_rate": 1.2903943077379188e-05,
+      "loss": 0.2591,
+      "step": 25100
+    },
+    {
+      "epoch": 59.29,
+      "learning_rate": 1.2830566261488289e-05,
+      "loss": 0.2606,
+      "step": 25200
+    },
+    {
+      "epoch": 59.53,
+      "learning_rate": 1.2756448265638897e-05,
+      "loss": 0.2581,
+      "step": 25300
+    },
+    {
+      "epoch": 59.76,
+      "learning_rate": 1.2682330269789505e-05,
+      "loss": 0.2574,
+      "step": 25400
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 1.2608212273940113e-05,
+      "loss": 0.2509,
+      "step": 25500
+    },
+    {
+      "epoch": 60.0,
+      "eval_loss": 0.36091676354408264,
+      "eval_runtime": 663.9043,
+      "eval_samples_per_second": 12.686,
+      "eval_wer": 0.2987519601871412,
+      "step": 25500
+    },
+    {
+      "epoch": 60.23,
+      "learning_rate": 1.2534094278090721e-05,
+      "loss": 0.2566,
+      "step": 25600
+    },
+    {
+      "epoch": 60.47,
+      "learning_rate": 1.2459976282241328e-05,
+      "loss": 0.2556,
+      "step": 25700
+    },
+    {
+      "epoch": 60.7,
+      "learning_rate": 1.2385858286391937e-05,
+      "loss": 0.2539,
+      "step": 25800
+    },
+    {
+      "epoch": 60.94,
+      "learning_rate": 1.2311740290542544e-05,
+      "loss": 0.2538,
+      "step": 25900
+    },
+    {
+      "epoch": 61.0,
+      "eval_loss": 0.3586573004722595,
+      "eval_runtime": 317.4988,
+      "eval_samples_per_second": 26.526,
+      "eval_wer": 0.29758556783867496,
+      "step": 25925
+    },
+    {
+      "epoch": 61.18,
+      "learning_rate": 1.2237622294693152e-05,
+      "loss": 0.2589,
+      "step": 26000
+    },
+    {
+      "epoch": 61.41,
+      "learning_rate": 1.2163504298843758e-05,
+      "loss": 0.249,
+      "step": 26100
+    },
+    {
+      "epoch": 61.65,
+      "learning_rate": 1.2089386302994368e-05,
+      "loss": 0.2464,
+      "step": 26200
+    },
+    {
+      "epoch": 61.88,
+      "learning_rate": 1.2015268307144976e-05,
+      "loss": 0.2563,
+      "step": 26300
+    },
+    {
+      "epoch": 62.0,
+      "eval_loss": 0.3534925878047943,
+      "eval_runtime": 317.8014,
+      "eval_samples_per_second": 26.501,
+      "eval_wer": 0.29697645183447596,
+      "step": 26350
+    },
+    {
+      "epoch": 62.12,
+      "learning_rate": 1.1941150311295582e-05,
+      "loss": 0.2559,
+      "step": 26400
+    },
+    {
+      "epoch": 62.35,
+      "learning_rate": 1.1867032315446192e-05,
+      "loss": 0.2543,
+      "step": 26500
+    },
+    {
+      "epoch": 62.59,
+      "learning_rate": 1.1792914319596798e-05,
+      "loss": 0.2496,
+      "step": 26600
+    },
+    {
+      "epoch": 62.82,
+      "learning_rate": 1.1718796323747406e-05,
+      "loss": 0.249,
+      "step": 26700
+    },
+    {
+      "epoch": 63.0,
+      "eval_loss": 0.34823477268218994,
+      "eval_runtime": 317.6983,
+      "eval_samples_per_second": 26.509,
+      "eval_wer": 0.29906299814673215,
+      "step": 26775
+    },
+    {
+      "epoch": 63.06,
+      "learning_rate": 1.1644678327898013e-05,
+      "loss": 0.2536,
+      "step": 26800
+    },
+    {
+      "epoch": 63.29,
+      "learning_rate": 1.1570560332048623e-05,
+      "loss": 0.2481,
+      "step": 26900
+    },
+    {
+      "epoch": 63.53,
+      "learning_rate": 1.1496442336199229e-05,
+      "loss": 0.2504,
+      "step": 27000
+    },
+    {
+      "epoch": 63.76,
+      "learning_rate": 1.1422324340349837e-05,
+      "loss": 0.2482,
+      "step": 27100
+    },
+    {
+      "epoch": 64.0,
+      "learning_rate": 1.1348206344500445e-05,
+      "loss": 0.2508,
+      "step": 27200
+    },
+    {
+      "epoch": 64.0,
+      "eval_loss": 0.35596856474876404,
+      "eval_runtime": 314.8406,
+      "eval_samples_per_second": 26.75,
+      "eval_wer": 0.29501950467204935,
+      "step": 27200
+    },
+    {
+      "epoch": 64.23,
+      "learning_rate": 1.1274088348651053e-05,
+      "loss": 0.2562,
+      "step": 27300
+    },
+    {
+      "epoch": 64.47,
+      "learning_rate": 1.1199970352801661e-05,
+      "loss": 0.2504,
+      "step": 27400
+    },
+    {
+      "epoch": 64.7,
+      "learning_rate": 1.1125852356952268e-05,
+      "loss": 0.25,
+      "step": 27500
+    },
+    {
+      "epoch": 64.94,
+      "learning_rate": 1.1051734361102877e-05,
+      "loss": 0.2518,
+      "step": 27600
+    },
+    {
+      "epoch": 65.0,
+      "eval_loss": 0.35186004638671875,
+      "eval_runtime": 317.5788,
+      "eval_samples_per_second": 26.519,
+      "eval_wer": 0.2965358147250554,
+      "step": 27625
+    },
+    {
+      "epoch": 65.18,
+      "learning_rate": 1.0977616365253484e-05,
+      "loss": 0.2502,
+      "step": 27700
+    },
+    {
+      "epoch": 65.41,
+      "learning_rate": 1.0903498369404092e-05,
+      "loss": 0.2451,
+      "step": 27800
+    },
+    {
+      "epoch": 65.65,
+      "learning_rate": 1.0829380373554698e-05,
+      "loss": 0.2508,
+      "step": 27900
+    },
+    {
+      "epoch": 65.88,
+      "learning_rate": 1.0755262377705308e-05,
+      "loss": 0.248,
+      "step": 28000
+    },
+    {
+      "epoch": 66.0,
+      "eval_loss": 0.34826791286468506,
+      "eval_runtime": 315.877,
+      "eval_samples_per_second": 26.662,
+      "eval_wer": 0.2940863907932764,
+      "step": 28050
+    },
+    {
+      "epoch": 66.12,
+      "learning_rate": 1.0681144381855916e-05,
+      "loss": 0.2488,
+      "step": 28100
+    },
+    {
+      "epoch": 66.35,
+      "learning_rate": 1.0607026386006522e-05,
+      "loss": 0.2446,
+      "step": 28200
+    },
+    {
+      "epoch": 66.59,
+      "learning_rate": 1.053290839015713e-05,
+      "loss": 0.2472,
+      "step": 28300
+    },
+    {
+      "epoch": 66.82,
+      "learning_rate": 1.0458790394307738e-05,
+      "loss": 0.2427,
+      "step": 28400
+    },
+    {
+      "epoch": 67.0,
+      "eval_loss": 0.35498106479644775,
+      "eval_runtime": 315.9906,
+      "eval_samples_per_second": 26.653,
+      "eval_wer": 0.29247936133538965,
+      "step": 28475
+    },
+    {
+      "epoch": 67.06,
+      "learning_rate": 1.0384672398458346e-05,
+      "loss": 0.2519,
+      "step": 28500
+    },
+    {
+      "epoch": 67.29,
+      "learning_rate": 1.0310554402608953e-05,
+      "loss": 0.2463,
+      "step": 28600
+    },
+    {
+      "epoch": 67.53,
+      "learning_rate": 1.0236436406759562e-05,
+      "loss": 0.2406,
+      "step": 28700
+    },
+    {
+      "epoch": 67.76,
+      "learning_rate": 1.0162318410910169e-05,
+      "loss": 0.2434,
+      "step": 28800
+    },
+    {
+      "epoch": 68.0,
+      "learning_rate": 1.0088200415060777e-05,
+      "loss": 0.2439,
+      "step": 28900
+    },
+    {
+      "epoch": 68.0,
+      "eval_loss": 0.35530757904052734,
+      "eval_runtime": 316.4428,
+      "eval_samples_per_second": 26.615,
+      "eval_wer": 0.2964710151501406,
+      "step": 28900
+    },
+    {
+      "epoch": 68.23,
+      "learning_rate": 1.0014082419211385e-05,
+      "loss": 0.2444,
+      "step": 29000
+    },
+    {
+      "epoch": 68.47,
+      "learning_rate": 9.939964423361993e-06,
+      "loss": 0.2459,
+      "step": 29100
+    },
+    {
+      "epoch": 68.7,
+      "learning_rate": 9.865846427512601e-06,
+      "loss": 0.2397,
+      "step": 29200
+    },
+    {
+      "epoch": 68.94,
+      "learning_rate": 9.791728431663207e-06,
+      "loss": 0.2427,
+      "step": 29300
+    },
+    {
+      "epoch": 69.0,
+      "eval_loss": 0.35256123542785645,
+      "eval_runtime": 316.7354,
+      "eval_samples_per_second": 26.59,
+      "eval_wer": 0.29481214603232203,
+      "step": 29325
+    },
+    {
+      "epoch": 69.18,
+      "learning_rate": 9.717610435813815e-06,
+      "loss": 0.2464,
+      "step": 29400
+    },
+    {
+      "epoch": 69.41,
+      "learning_rate": 9.643492439964424e-06,
+      "loss": 0.2396,
+      "step": 29500
+    },
+    {
+      "epoch": 69.65,
+      "learning_rate": 9.569374444115032e-06,
+      "loss": 0.2441,
+      "step": 29600
+    },
+    {
+      "epoch": 69.88,
+      "learning_rate": 9.495256448265638e-06,
+      "loss": 0.2412,
+      "step": 29700
+    },
+    {
+      "epoch": 70.0,
+      "eval_loss": 0.356432169675827,
+      "eval_runtime": 316.2059,
+      "eval_samples_per_second": 26.635,
+      "eval_wer": 0.2973004497090499,
+      "step": 29750
+    },
+    {
+      "epoch": 70.12,
+      "learning_rate": 9.421879632374742e-06,
+      "loss": 0.2488,
+      "step": 29800
+    },
+    {
+      "epoch": 70.35,
+      "learning_rate": 9.347761636525348e-06,
+      "loss": 0.2348,
+      "step": 29900
+    },
+    {
+      "epoch": 70.59,
+      "learning_rate": 9.273643640675956e-06,
+      "loss": 0.2426,
+      "step": 30000
+    },
+    {
+      "epoch": 70.82,
+      "learning_rate": 9.199525644826563e-06,
+      "loss": 0.2394,
+      "step": 30100
+    },
+    {
+      "epoch": 71.0,
+      "eval_loss": 0.3529145419597626,
+      "eval_runtime": 314.7426,
+      "eval_samples_per_second": 26.758,
+      "eval_wer": 0.28974481927398554,
+      "step": 30175
+    },
+    {
+      "epoch": 71.06,
+      "learning_rate": 9.125407648977172e-06,
+      "loss": 0.2451,
+      "step": 30200
+    },
+    {
+      "epoch": 71.29,
+      "learning_rate": 9.051289653127779e-06,
+      "loss": 0.2371,
+      "step": 30300
+    },
+    {
+      "epoch": 71.53,
+      "learning_rate": 8.977171657278387e-06,
+      "loss": 0.2367,
+      "step": 30400
+    },
+    {
+      "epoch": 71.76,
+      "learning_rate": 8.903053661428997e-06,
+      "loss": 0.2417,
+      "step": 30500
+    },
+    {
+      "epoch": 72.0,
+      "learning_rate": 8.828935665579603e-06,
+      "loss": 0.247,
+      "step": 30600
+    },
+    {
+      "epoch": 72.0,
+      "eval_loss": 0.3524751663208008,
+      "eval_runtime": 316.256,
+      "eval_samples_per_second": 26.63,
+      "eval_wer": 0.2913777685618382,
+      "step": 30600
+    },
+    {
+      "epoch": 72.23,
+      "learning_rate": 8.754817669730211e-06,
+      "loss": 0.2448,
+      "step": 30700
+    },
+    {
+      "epoch": 72.47,
+      "learning_rate": 8.680699673880817e-06,
+      "loss": 0.238,
+      "step": 30800
+    },
+    {
+      "epoch": 72.7,
+      "learning_rate": 8.606581678031427e-06,
+      "loss": 0.2341,
+      "step": 30900
+    },
+    {
+      "epoch": 72.94,
+      "learning_rate": 8.532463682182034e-06,
+      "loss": 0.2371,
+      "step": 31000
+    },
+    {
+      "epoch": 73.0,
+      "eval_loss": 0.3488871455192566,
+      "eval_runtime": 316.856,
+      "eval_samples_per_second": 26.58,
+      "eval_wer": 0.28990033825378103,
+      "step": 31025
+    },
+    {
+      "epoch": 73.0,
+      "step": 31025,
+      "total_flos": 0,
+      "train_runtime": 465541.7438,
+      "train_samples_per_second": 0.091
+    }
+  ],
+  "max_steps": 42500,
+  "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 sha256:0ff4c761c4b0ff5e4df7b294a0a4dc6efaa59dc7be6ae8a4850f25dc2ceaecc3
+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}