Add checkpoints

all_results.json

@@ -0,0 +1,23 @@
+{
+    "epoch": 79.0,
+    "eval_loss": 0.33847060799598694,
+    "eval_mem_cpu_alloc_delta": -354455552,
+    "eval_mem_cpu_peaked_delta": 406319104,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 5722557440,
+    "eval_runtime": 135.2302,
+    "eval_samples": 1500,
+    "eval_samples_per_second": 11.092,
+    "eval_wer": 0.19016709793430112,
+    "init_mem_cpu_alloc_delta": 1954693120,
+    "init_mem_cpu_peaked_delta": 0,
+    "init_mem_gpu_alloc_delta": 1261939712,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 1794605056,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3804792832,
+    "train_mem_gpu_peaked_delta": 9913940480,
+    "train_runtime": 205971.9249,
+    "train_samples": 30823,
+    "train_samples_per_second": 0.109
+}

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": 45
+}

config_train.json

@@ -0,0 +1,190 @@
+{
+    "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.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": 140,                // 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": 0.06666666667, // 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/PT/140-epoch/",
+    // CACHE
+    "dataset_cache": "../datasets/",
+
+    // DATASETS
+    "datasets":{ 
+        
+        "files_path": "/workspace/scratch/ecasanova/datasets/YourTTS2ASR/Common_Voice/cv-corpus-7.0-2021-07-21/pt/", // 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/pt/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/TTS-Portuguese-Corpus_16khz/train_TTS-Portuguese_Corpus_metadata_converted_to_ASR_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/TTS-Portuguese-Corpus_16khz/train_TTS-Portuguese_Corpus_metadata_converted_to_ASR.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/pt/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/TTS-Portuguese-Corpus_16khz/eval_TTS-Portuguese_Corpus_metadata_converted_to_ASR_copy_generated_VC_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/TTS-Portuguese-Corpus_16khz/eval_TTS-Portuguese_Corpus_metadata_converted_to_ASR.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/pt/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": 79.0,
+    "eval_loss": 0.33847060799598694,
+    "eval_mem_cpu_alloc_delta": -354455552,
+    "eval_mem_cpu_peaked_delta": 406319104,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 5722557440,
+    "eval_runtime": 135.2302,
+    "eval_samples": 1500,
+    "eval_samples_per_second": 11.092,
+    "eval_wer": 0.19016709793430112
+}

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:eceb36972fbbe269122407bdcd2443cc8486ed88d986ecf5bd49678ee3c0c4e2
+size 1262118359

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": 79.0,
+    "init_mem_cpu_alloc_delta": 1954693120,
+    "init_mem_cpu_peaked_delta": 0,
+    "init_mem_gpu_alloc_delta": 1261939712,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 1794605056,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3804792832,
+    "train_mem_gpu_peaked_delta": 9913940480,
+    "train_runtime": 205971.9249,
+    "train_samples": 30823,
+    "train_samples_per_second": 0.109
+}

trainer_state.json

@@ -0,0 +1,1417 @@
+{
+  "best_metric": 0.3218751549720764,
+  "best_model_checkpoint": "../checkpoints/Wav2Vec-voxpopuli/one-speaker/Final-paper/GEN/PT/140-epoch/checkpoint-11040",
+  "epoch": 78.99662600570984,
+  "global_step": 12640,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.008032128514056e-08,
+      "loss": 12.1375,
+      "step": 1
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 2.0080321285140564e-06,
+      "loss": 12.2606,
+      "step": 100
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 11.042320251464844,
+      "eval_runtime": 128.6432,
+      "eval_samples_per_second": 11.66,
+      "eval_wer": 1.008201394675672,
+      "step": 160
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 4.016064257028113e-06,
+      "loss": 11.2029,
+      "step": 200
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 6.024096385542168e-06,
+      "loss": 8.2977,
+      "step": 300
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 6.322738170623779,
+      "eval_runtime": 127.7737,
+      "eval_samples_per_second": 11.74,
+      "eval_wer": 1.0,
+      "step": 320
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 8.032128514056226e-06,
+      "loss": 6.4999,
+      "step": 400
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 4.472578525543213,
+      "eval_runtime": 134.2762,
+      "eval_samples_per_second": 11.171,
+      "eval_wer": 1.0,
+      "step": 480
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 1.0040160642570281e-05,
+      "loss": 5.3252,
+      "step": 500
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 1.2048192771084337e-05,
+      "loss": 4.4851,
+      "step": 600
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 3.6344993114471436,
+      "eval_runtime": 133.8576,
+      "eval_samples_per_second": 11.206,
+      "eval_wer": 1.0,
+      "step": 640
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 1.4056224899598394e-05,
+      "loss": 3.8872,
+      "step": 700
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 1.606425702811245e-05,
+      "loss": 3.4357,
+      "step": 800
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 3.175006628036499,
+      "eval_runtime": 134.2322,
+      "eval_samples_per_second": 11.175,
+      "eval_wer": 1.0,
+      "step": 800
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 1.807228915662651e-05,
+      "loss": 3.1686,
+      "step": 900
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 2.968130111694336,
+      "eval_runtime": 134.7282,
+      "eval_samples_per_second": 11.134,
+      "eval_wer": 1.0,
+      "step": 960
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 2.0080321285140562e-05,
+      "loss": 3.0199,
+      "step": 1000
+    },
+    {
+      "epoch": 6.87,
+      "learning_rate": 2.208835341365462e-05,
+      "loss": 2.9408,
+      "step": 1100
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 2.8903684616088867,
+      "eval_runtime": 134.0645,
+      "eval_samples_per_second": 11.189,
+      "eval_wer": 1.0,
+      "step": 1120
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 2.4096385542168674e-05,
+      "loss": 2.9099,
+      "step": 1200
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 2.7165446281433105,
+      "eval_runtime": 133.6155,
+      "eval_samples_per_second": 11.226,
+      "eval_wer": 0.9993421341169247,
+      "step": 1280
+    },
+    {
+      "epoch": 8.12,
+      "learning_rate": 2.6104417670682734e-05,
+      "loss": 2.8137,
+      "step": 1300
+    },
+    {
+      "epoch": 8.75,
+      "learning_rate": 2.8112449799196788e-05,
+      "loss": 2.6131,
+      "step": 1400
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 2.137465476989746,
+      "eval_runtime": 132.9892,
+      "eval_samples_per_second": 11.279,
+      "eval_wer": 0.9312310863558616,
+      "step": 1440
+    },
+    {
+      "epoch": 9.37,
+      "learning_rate": 2.9991390031569885e-05,
+      "loss": 2.2614,
+      "step": 1500
+    },
+    {
+      "epoch": 10.0,
+      "learning_rate": 2.9847890557734623e-05,
+      "loss": 1.77,
+      "step": 1600
+    },
+    {
+      "epoch": 10.0,
+      "eval_loss": 1.2687721252441406,
+      "eval_runtime": 133.0804,
+      "eval_samples_per_second": 11.271,
+      "eval_wer": 0.7300118415858954,
+      "step": 1600
+    },
+    {
+      "epoch": 10.62,
+      "learning_rate": 2.9704391083899357e-05,
+      "loss": 1.4261,
+      "step": 1700
+    },
+    {
+      "epoch": 11.0,
+      "eval_loss": 0.9103694558143616,
+      "eval_runtime": 132.9746,
+      "eval_samples_per_second": 11.28,
+      "eval_wer": 0.5371255646682163,
+      "step": 1760
+    },
+    {
+      "epoch": 11.25,
+      "learning_rate": 2.9560891610064095e-05,
+      "loss": 1.2212,
+      "step": 1800
+    },
+    {
+      "epoch": 11.87,
+      "learning_rate": 2.9417392136228833e-05,
+      "loss": 1.075,
+      "step": 1900
+    },
+    {
+      "epoch": 12.0,
+      "eval_loss": 0.7756706476211548,
+      "eval_runtime": 132.9148,
+      "eval_samples_per_second": 11.285,
+      "eval_wer": 0.4647164598043945,
+      "step": 1920
+    },
+    {
+      "epoch": 12.5,
+      "learning_rate": 2.927389266239357e-05,
+      "loss": 0.9948,
+      "step": 2000
+    },
+    {
+      "epoch": 13.0,
+      "eval_loss": 0.666081964969635,
+      "eval_runtime": 134.3606,
+      "eval_samples_per_second": 11.164,
+      "eval_wer": 0.4066488311916144,
+      "step": 2080
+    },
+    {
+      "epoch": 13.12,
+      "learning_rate": 2.9130393188558308e-05,
+      "loss": 0.9185,
+      "step": 2100
+    },
+    {
+      "epoch": 13.75,
+      "learning_rate": 2.8986893714723046e-05,
+      "loss": 0.8638,
+      "step": 2200
+    },
+    {
+      "epoch": 14.0,
+      "eval_loss": 0.6328504085540771,
+      "eval_runtime": 134.3957,
+      "eval_samples_per_second": 11.161,
+      "eval_wer": 0.37897460637691327,
+      "step": 2240
+    },
+    {
+      "epoch": 14.37,
+      "learning_rate": 2.8843394240887784e-05,
+      "loss": 0.8171,
+      "step": 2300
+    },
+    {
+      "epoch": 15.0,
+      "learning_rate": 2.869989476705252e-05,
+      "loss": 0.7853,
+      "step": 2400
+    },
+    {
+      "epoch": 15.0,
+      "eval_loss": 0.5940719246864319,
+      "eval_runtime": 134.4528,
+      "eval_samples_per_second": 11.156,
+      "eval_wer": 0.3550721459585106,
+      "step": 2400
+    },
+    {
+      "epoch": 15.62,
+      "learning_rate": 2.855639529321726e-05,
+      "loss": 0.753,
+      "step": 2500
+    },
+    {
+      "epoch": 16.0,
+      "eval_loss": 0.5752812623977661,
+      "eval_runtime": 135.1505,
+      "eval_samples_per_second": 11.099,
+      "eval_wer": 0.34020437700100875,
+      "step": 2560
+    },
+    {
+      "epoch": 16.25,
+      "learning_rate": 2.8412895819381997e-05,
+      "loss": 0.7261,
+      "step": 2600
+    },
+    {
+      "epoch": 16.87,
+      "learning_rate": 2.8269396345546735e-05,
+      "loss": 0.7005,
+      "step": 2700
+    },
+    {
+      "epoch": 17.0,
+      "eval_loss": 0.512348473072052,
+      "eval_runtime": 134.1054,
+      "eval_samples_per_second": 11.185,
+      "eval_wer": 0.3196351037235209,
+      "step": 2720
+    },
+    {
+      "epoch": 17.5,
+      "learning_rate": 2.8125896871711473e-05,
+      "loss": 0.6796,
+      "step": 2800
+    },
+    {
+      "epoch": 18.0,
+      "eval_loss": 0.5050208568572998,
+      "eval_runtime": 134.5141,
+      "eval_samples_per_second": 11.151,
+      "eval_wer": 0.3070917942195518,
+      "step": 2880
+    },
+    {
+      "epoch": 18.12,
+      "learning_rate": 2.798239739787621e-05,
+      "loss": 0.6586,
+      "step": 2900
+    },
+    {
+      "epoch": 18.75,
+      "learning_rate": 2.7838897924040948e-05,
+      "loss": 0.6396,
+      "step": 3000
+    },
+    {
+      "epoch": 19.0,
+      "eval_loss": 0.5050458312034607,
+      "eval_runtime": 134.3675,
+      "eval_samples_per_second": 11.163,
+      "eval_wer": 0.296697513266962,
+      "step": 3040
+    },
+    {
+      "epoch": 19.37,
+      "learning_rate": 2.7695398450205686e-05,
+      "loss": 0.6189,
+      "step": 3100
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 2.7551898976370424e-05,
+      "loss": 0.6127,
+      "step": 3200
+    },
+    {
+      "epoch": 20.0,
+      "eval_loss": 0.46447858214378357,
+      "eval_runtime": 133.8616,
+      "eval_samples_per_second": 11.206,
+      "eval_wer": 0.28691724047190914,
+      "step": 3200
+    },
+    {
+      "epoch": 20.62,
+      "learning_rate": 2.7408399502535158e-05,
+      "loss": 0.5993,
+      "step": 3300
+    },
+    {
+      "epoch": 21.0,
+      "eval_loss": 0.4493682384490967,
+      "eval_runtime": 134.4621,
+      "eval_samples_per_second": 11.156,
+      "eval_wer": 0.28011929301346433,
+      "step": 3360
+    },
+    {
+      "epoch": 21.25,
+      "learning_rate": 2.7264900028699892e-05,
+      "loss": 0.5843,
+      "step": 3400
+    },
+    {
+      "epoch": 21.87,
+      "learning_rate": 2.712140055486463e-05,
+      "loss": 0.58,
+      "step": 3500
+    },
+    {
+      "epoch": 22.0,
+      "eval_loss": 0.44041261076927185,
+      "eval_runtime": 134.062,
+      "eval_samples_per_second": 11.189,
+      "eval_wer": 0.2688039998245691,
+      "step": 3520
+    },
+    {
+      "epoch": 22.5,
+      "learning_rate": 2.6977901081029368e-05,
+      "loss": 0.5584,
+      "step": 3600
+    },
+    {
+      "epoch": 23.0,
+      "eval_loss": 0.45153650641441345,
+      "eval_runtime": 133.1081,
+      "eval_samples_per_second": 11.269,
+      "eval_wer": 0.27783869128546995,
+      "step": 3680
+    },
+    {
+      "epoch": 23.12,
+      "learning_rate": 2.6834401607194106e-05,
+      "loss": 0.5549,
+      "step": 3700
+    },
+    {
+      "epoch": 23.75,
+      "learning_rate": 2.6690902133358844e-05,
+      "loss": 0.5425,
+      "step": 3800
+    },
+    {
+      "epoch": 24.0,
+      "eval_loss": 0.42425668239593506,
+      "eval_runtime": 133.8644,
+      "eval_samples_per_second": 11.205,
+      "eval_wer": 0.2665233980965747,
+      "step": 3840
+    },
+    {
+      "epoch": 24.37,
+      "learning_rate": 2.654740265952358e-05,
+      "loss": 0.531,
+      "step": 3900
+    },
+    {
+      "epoch": 25.0,
+      "learning_rate": 2.640390318568832e-05,
+      "loss": 0.5273,
+      "step": 4000
+    },
+    {
+      "epoch": 25.0,
+      "eval_loss": 0.43647119402885437,
+      "eval_runtime": 132.965,
+      "eval_samples_per_second": 11.281,
+      "eval_wer": 0.26696197535195826,
+      "step": 4000
+    },
+    {
+      "epoch": 25.62,
+      "learning_rate": 2.6260403711853057e-05,
+      "loss": 0.5147,
+      "step": 4100
+    },
+    {
+      "epoch": 26.0,
+      "eval_loss": 0.41047143936157227,
+      "eval_runtime": 135.0829,
+      "eval_samples_per_second": 11.104,
+      "eval_wer": 0.24959431603877022,
+      "step": 4160
+    },
+    {
+      "epoch": 26.25,
+      "learning_rate": 2.6116904238017795e-05,
+      "loss": 0.5136,
+      "step": 4200
+    },
+    {
+      "epoch": 26.87,
+      "learning_rate": 2.5973404764182532e-05,
+      "loss": 0.5042,
+      "step": 4300
+    },
+    {
+      "epoch": 27.0,
+      "eval_loss": 0.4049687385559082,
+      "eval_runtime": 134.3457,
+      "eval_samples_per_second": 11.165,
+      "eval_wer": 0.2538046576904522,
+      "step": 4320
+    },
+    {
+      "epoch": 27.5,
+      "learning_rate": 2.582990529034727e-05,
+      "loss": 0.4925,
+      "step": 4400
+    },
+    {
+      "epoch": 28.0,
+      "eval_loss": 0.41241469979286194,
+      "eval_runtime": 134.126,
+      "eval_samples_per_second": 11.184,
+      "eval_wer": 0.251173194158151,
+      "step": 4480
+    },
+    {
+      "epoch": 28.12,
+      "learning_rate": 2.5686405816512008e-05,
+      "loss": 0.4925,
+      "step": 4500
+    },
+    {
+      "epoch": 28.75,
+      "learning_rate": 2.5542906342676746e-05,
+      "loss": 0.4753,
+      "step": 4600
+    },
+    {
+      "epoch": 29.0,
+      "eval_loss": 0.40935131907463074,
+      "eval_runtime": 134.4907,
+      "eval_samples_per_second": 11.153,
+      "eval_wer": 0.24437524669970614,
+      "step": 4640
+    },
+    {
+      "epoch": 29.37,
+      "learning_rate": 2.5399406868841483e-05,
+      "loss": 0.4773,
+      "step": 4700
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 2.525590739500622e-05,
+      "loss": 0.4706,
+      "step": 4800
+    },
+    {
+      "epoch": 30.0,
+      "eval_loss": 0.3854382336139679,
+      "eval_runtime": 134.5082,
+      "eval_samples_per_second": 11.152,
+      "eval_wer": 0.23424411210034649,
+      "step": 4800
+    },
+    {
+      "epoch": 30.62,
+      "learning_rate": 2.511240792117096e-05,
+      "loss": 0.4625,
+      "step": 4900
+    },
+    {
+      "epoch": 31.0,
+      "eval_loss": 0.3864579200744629,
+      "eval_runtime": 134.0786,
+      "eval_samples_per_second": 11.187,
+      "eval_wer": 0.23937546598833384,
+      "step": 4960
+    },
+    {
+      "epoch": 31.25,
+      "learning_rate": 2.4968908447335693e-05,
+      "loss": 0.4586,
+      "step": 5000
+    },
+    {
+      "epoch": 31.87,
+      "learning_rate": 2.482540897350043e-05,
+      "loss": 0.4512,
+      "step": 5100
+    },
+    {
+      "epoch": 32.0,
+      "eval_loss": 0.3823211193084717,
+      "eval_runtime": 134.3959,
+      "eval_samples_per_second": 11.161,
+      "eval_wer": 0.2313056444892768,
+      "step": 5120
+    },
+    {
+      "epoch": 32.5,
+      "learning_rate": 2.468190949966517e-05,
+      "loss": 0.4445,
+      "step": 5200
+    },
+    {
+      "epoch": 33.0,
+      "eval_loss": 0.38146257400512695,
+      "eval_runtime": 134.7806,
+      "eval_samples_per_second": 11.129,
+      "eval_wer": 0.23472654708126836,
+      "step": 5280
+    },
+    {
+      "epoch": 33.12,
+      "learning_rate": 2.4538410025829903e-05,
+      "loss": 0.4451,
+      "step": 5300
+    },
+    {
+      "epoch": 33.75,
+      "learning_rate": 2.439491055199464e-05,
+      "loss": 0.435,
+      "step": 5400
+    },
+    {
+      "epoch": 34.0,
+      "eval_loss": 0.3864315450191498,
+      "eval_runtime": 134.161,
+      "eval_samples_per_second": 11.181,
+      "eval_wer": 0.23200736809789044,
+      "step": 5440
+    },
+    {
+      "epoch": 34.37,
+      "learning_rate": 2.425141107815938e-05,
+      "loss": 0.4342,
+      "step": 5500
+    },
+    {
+      "epoch": 35.0,
+      "learning_rate": 2.4107911604324117e-05,
+      "loss": 0.4284,
+      "step": 5600
+    },
+    {
+      "epoch": 35.0,
+      "eval_loss": 0.370952308177948,
+      "eval_runtime": 134.1583,
+      "eval_samples_per_second": 11.181,
+      "eval_wer": 0.22547256699267576,
+      "step": 5600
+    },
+    {
+      "epoch": 35.62,
+      "learning_rate": 2.3964412130488854e-05,
+      "loss": 0.4271,
+      "step": 5700
+    },
+    {
+      "epoch": 36.0,
+      "eval_loss": 0.3921850621700287,
+      "eval_runtime": 133.9964,
+      "eval_samples_per_second": 11.194,
+      "eval_wer": 0.2336739616683479,
+      "step": 5760
+    },
+    {
+      "epoch": 36.25,
+      "learning_rate": 2.3820912656653592e-05,
+      "loss": 0.4137,
+      "step": 5800
+    },
+    {
+      "epoch": 36.87,
+      "learning_rate": 2.367741318281833e-05,
+      "loss": 0.4143,
+      "step": 5900
+    },
+    {
+      "epoch": 37.0,
+      "eval_loss": 0.37270987033843994,
+      "eval_runtime": 133.8613,
+      "eval_samples_per_second": 11.206,
+      "eval_wer": 0.2222270953028376,
+      "step": 5920
+    },
+    {
+      "epoch": 37.5,
+      "learning_rate": 2.3533913708983068e-05,
+      "loss": 0.4125,
+      "step": 6000
+    },
+    {
+      "epoch": 38.0,
+      "eval_loss": 0.37099114060401917,
+      "eval_runtime": 133.4063,
+      "eval_samples_per_second": 11.244,
+      "eval_wer": 0.21977106267268978,
+      "step": 6080
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 2.3390414235147805e-05,
+      "loss": 0.4025,
+      "step": 6100
+    },
+    {
+      "epoch": 38.75,
+      "learning_rate": 2.3246914761312543e-05,
+      "loss": 0.4011,
+      "step": 6200
+    },
+    {
+      "epoch": 39.0,
+      "eval_loss": 0.36733388900756836,
+      "eval_runtime": 134.2886,
+      "eval_samples_per_second": 11.17,
+      "eval_wer": 0.2247708433840621,
+      "step": 6240
+    },
+    {
+      "epoch": 39.37,
+      "learning_rate": 2.310341528747728e-05,
+      "loss": 0.395,
+      "step": 6300
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 2.295991581364202e-05,
+      "loss": 0.3976,
+      "step": 6400
+    },
+    {
+      "epoch": 40.0,
+      "eval_loss": 0.38662710785865784,
+      "eval_runtime": 134.9794,
+      "eval_samples_per_second": 11.113,
+      "eval_wer": 0.22384983114775667,
+      "step": 6400
+    },
+    {
+      "epoch": 40.62,
+      "learning_rate": 2.2816416339806756e-05,
+      "loss": 0.3912,
+      "step": 6500
+    },
+    {
+      "epoch": 41.0,
+      "eval_loss": 0.368117094039917,
+      "eval_runtime": 134.9215,
+      "eval_samples_per_second": 11.118,
+      "eval_wer": 0.22082364808561028,
+      "step": 6560
+    },
+    {
+      "epoch": 41.25,
+      "learning_rate": 2.2672916865971494e-05,
+      "loss": 0.3841,
+      "step": 6600
+    },
+    {
+      "epoch": 41.87,
+      "learning_rate": 2.252941739213623e-05,
+      "loss": 0.3856,
+      "step": 6700
+    },
+    {
+      "epoch": 42.0,
+      "eval_loss": 0.34648868441581726,
+      "eval_runtime": 135.1982,
+      "eval_samples_per_second": 11.095,
+      "eval_wer": 0.21569229419762292,
+      "step": 6720
+    },
+    {
+      "epoch": 42.5,
+      "learning_rate": 2.2385917918300966e-05,
+      "loss": 0.3814,
+      "step": 6800
+    },
+    {
+      "epoch": 43.0,
+      "eval_loss": 0.35718342661857605,
+      "eval_runtime": 134.5426,
+      "eval_samples_per_second": 11.149,
+      "eval_wer": 0.2124906802333231,
+      "step": 6880
+    },
+    {
+      "epoch": 43.12,
+      "learning_rate": 2.2242418444465704e-05,
+      "loss": 0.3765,
+      "step": 6900
+    },
+    {
+      "epoch": 43.75,
+      "learning_rate": 2.2098918970630442e-05,
+      "loss": 0.3731,
+      "step": 7000
+    },
+    {
+      "epoch": 44.0,
+      "eval_loss": 0.3691784143447876,
+      "eval_runtime": 134.9182,
+      "eval_samples_per_second": 11.118,
+      "eval_wer": 0.21021007850532872,
+      "step": 7040
+    },
+    {
+      "epoch": 44.37,
+      "learning_rate": 2.195541949679518e-05,
+      "loss": 0.3679,
+      "step": 7100
+    },
+    {
+      "epoch": 45.0,
+      "learning_rate": 2.1811920022959917e-05,
+      "loss": 0.363,
+      "step": 7200
+    },
+    {
+      "epoch": 45.0,
+      "eval_loss": 0.3651253283023834,
+      "eval_runtime": 136.1991,
+      "eval_samples_per_second": 11.013,
+      "eval_wer": 0.21183281435024778,
+      "step": 7200
+    },
+    {
+      "epoch": 45.62,
+      "learning_rate": 2.1668420549124652e-05,
+      "loss": 0.3628,
+      "step": 7300
+    },
+    {
+      "epoch": 46.0,
+      "eval_loss": 0.35606908798217773,
+      "eval_runtime": 136.1343,
+      "eval_samples_per_second": 11.019,
+      "eval_wer": 0.20889434673917812,
+      "step": 7360
+    },
+    {
+      "epoch": 46.25,
+      "learning_rate": 2.152492107528939e-05,
+      "loss": 0.364,
+      "step": 7400
+    },
+    {
+      "epoch": 46.87,
+      "learning_rate": 2.1381421601454127e-05,
+      "loss": 0.3582,
+      "step": 7500
+    },
+    {
+      "epoch": 47.0,
+      "eval_loss": 0.3635926842689514,
+      "eval_runtime": 136.0098,
+      "eval_samples_per_second": 11.029,
+      "eval_wer": 0.20714003771764397,
+      "step": 7520
+    },
+    {
+      "epoch": 47.5,
+      "learning_rate": 2.1237922127618865e-05,
+      "loss": 0.3563,
+      "step": 7600
+    },
+    {
+      "epoch": 48.0,
+      "eval_loss": 0.3590168356895447,
+      "eval_runtime": 136.0862,
+      "eval_samples_per_second": 11.022,
+      "eval_wer": 0.21227139160563133,
+      "step": 7680
+    },
+    {
+      "epoch": 48.12,
+      "learning_rate": 2.1094422653783603e-05,
+      "loss": 0.3535,
+      "step": 7700
+    },
+    {
+      "epoch": 48.75,
+      "learning_rate": 2.095092317994834e-05,
+      "loss": 0.3518,
+      "step": 7800
+    },
+    {
+      "epoch": 49.0,
+      "eval_loss": 0.3534136414527893,
+      "eval_runtime": 135.7077,
+      "eval_samples_per_second": 11.053,
+      "eval_wer": 0.2056050173238016,
+      "step": 7840
+    },
+    {
+      "epoch": 49.37,
+      "learning_rate": 2.080742370611308e-05,
+      "loss": 0.3483,
+      "step": 7900
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 2.0663924232277816e-05,
+      "loss": 0.3482,
+      "step": 8000
+    },
+    {
+      "epoch": 50.0,
+      "eval_loss": 0.35351213812828064,
+      "eval_runtime": 135.799,
+      "eval_samples_per_second": 11.046,
+      "eval_wer": 0.2045524319108811,
+      "step": 8000
+    },
+    {
+      "epoch": 50.62,
+      "learning_rate": 2.0521859753180906e-05,
+      "loss": 0.345,
+      "step": 8100
+    },
+    {
+      "epoch": 51.0,
+      "eval_loss": 0.35218343138694763,
+      "eval_runtime": 137.1629,
+      "eval_samples_per_second": 10.936,
+      "eval_wer": 0.20389456602780578,
+      "step": 8160
+    },
+    {
+      "epoch": 51.25,
+      "learning_rate": 2.037836027934564e-05,
+      "loss": 0.3384,
+      "step": 8200
+    },
+    {
+      "epoch": 51.87,
+      "learning_rate": 2.0234860805510378e-05,
+      "loss": 0.3379,
+      "step": 8300
+    },
+    {
+      "epoch": 52.0,
+      "eval_loss": 0.3376038074493408,
+      "eval_runtime": 135.192,
+      "eval_samples_per_second": 11.095,
+      "eval_wer": 0.20165782202534976,
+      "step": 8320
+    },
+    {
+      "epoch": 52.5,
+      "learning_rate": 2.0091361331675116e-05,
+      "loss": 0.3324,
+      "step": 8400
+    },
+    {
+      "epoch": 53.0,
+      "eval_loss": 0.349118173122406,
+      "eval_runtime": 136.9229,
+      "eval_samples_per_second": 10.955,
+      "eval_wer": 0.20428928555765097,
+      "step": 8480
+    },
+    {
+      "epoch": 53.12,
+      "learning_rate": 1.9947861857839854e-05,
+      "loss": 0.3352,
+      "step": 8500
+    },
+    {
+      "epoch": 53.75,
+      "learning_rate": 1.980436238400459e-05,
+      "loss": 0.3329,
+      "step": 8600
+    },
+    {
+      "epoch": 54.0,
+      "eval_loss": 0.3493926525115967,
+      "eval_runtime": 135.7895,
+      "eval_samples_per_second": 11.047,
+      "eval_wer": 0.20196482610411823,
+      "step": 8640
+    },
+    {
+      "epoch": 54.37,
+      "learning_rate": 1.966086291016933e-05,
+      "loss": 0.3264,
+      "step": 8700
+    },
+    {
+      "epoch": 55.0,
+      "learning_rate": 1.9517363436334067e-05,
+      "loss": 0.3303,
+      "step": 8800
+    },
+    {
+      "epoch": 55.0,
+      "eval_loss": 0.3516405522823334,
+      "eval_runtime": 134.0631,
+      "eval_samples_per_second": 11.189,
+      "eval_wer": 0.19902635849304856,
+      "step": 8800
+    },
+    {
+      "epoch": 55.62,
+      "learning_rate": 1.9373863962498805e-05,
+      "loss": 0.3258,
+      "step": 8900
+    },
+    {
+      "epoch": 56.0,
+      "eval_loss": 0.3470868468284607,
+      "eval_runtime": 134.5501,
+      "eval_samples_per_second": 11.148,
+      "eval_wer": 0.20385070830226745,
+      "step": 8960
+    },
+    {
+      "epoch": 56.25,
+      "learning_rate": 1.9230364488663542e-05,
+      "loss": 0.3261,
+      "step": 9000
+    },
+    {
+      "epoch": 56.87,
+      "learning_rate": 1.908686501482828e-05,
+      "loss": 0.3189,
+      "step": 9100
+    },
+    {
+      "epoch": 57.0,
+      "eval_loss": 0.3370627164840698,
+      "eval_runtime": 134.7548,
+      "eval_samples_per_second": 11.131,
+      "eval_wer": 0.20007894390596903,
+      "step": 9120
+    },
+    {
+      "epoch": 57.5,
+      "learning_rate": 1.8943365540993018e-05,
+      "loss": 0.3137,
+      "step": 9200
+    },
+    {
+      "epoch": 58.0,
+      "eval_loss": 0.34424567222595215,
+      "eval_runtime": 135.5794,
+      "eval_samples_per_second": 11.064,
+      "eval_wer": 0.20196482610411823,
+      "step": 9280
+    },
+    {
+      "epoch": 58.12,
+      "learning_rate": 1.8799866067157756e-05,
+      "loss": 0.3187,
+      "step": 9300
+    },
+    {
+      "epoch": 58.75,
+      "learning_rate": 1.8656366593322493e-05,
+      "loss": 0.3166,
+      "step": 9400
+    },
+    {
+      "epoch": 59.0,
+      "eval_loss": 0.347896009683609,
+      "eval_runtime": 136.6913,
+      "eval_samples_per_second": 10.974,
+      "eval_wer": 0.20288583834042367,
+      "step": 9440
+    },
+    {
+      "epoch": 59.37,
+      "learning_rate": 1.851286711948723e-05,
+      "loss": 0.3135,
+      "step": 9500
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 1.836936764565197e-05,
+      "loss": 0.3128,
+      "step": 9600
+    },
+    {
+      "epoch": 60.0,
+      "eval_loss": 0.3388006091117859,
+      "eval_runtime": 135.2694,
+      "eval_samples_per_second": 11.089,
+      "eval_wer": 0.1940704355072146,
+      "step": 9600
+    },
+    {
+      "epoch": 60.62,
+      "learning_rate": 1.8225868171816707e-05,
+      "loss": 0.3073,
+      "step": 9700
+    },
+    {
+      "epoch": 61.0,
+      "eval_loss": 0.3377033472061157,
+      "eval_runtime": 135.4444,
+      "eval_samples_per_second": 11.075,
+      "eval_wer": 0.19205298013245034,
+      "step": 9760
+    },
+    {
+      "epoch": 61.25,
+      "learning_rate": 1.808236869798144e-05,
+      "loss": 0.3061,
+      "step": 9800
+    },
+    {
+      "epoch": 61.87,
+      "learning_rate": 1.794030421888453e-05,
+      "loss": 0.3053,
+      "step": 9900
+    },
+    {
+      "epoch": 62.0,
+      "eval_loss": 0.3457684814929962,
+      "eval_runtime": 134.2365,
+      "eval_samples_per_second": 11.174,
+      "eval_wer": 0.19551774044998027,
+      "step": 9920
+    },
+    {
+      "epoch": 62.5,
+      "learning_rate": 1.779680474504927e-05,
+      "loss": 0.3021,
+      "step": 10000
+    },
+    {
+      "epoch": 63.0,
+      "eval_loss": 0.33506670594215393,
+      "eval_runtime": 135.3785,
+      "eval_samples_per_second": 11.08,
+      "eval_wer": 0.19490373229244332,
+      "step": 10080
+    },
+    {
+      "epoch": 63.12,
+      "learning_rate": 1.7653305271214006e-05,
+      "loss": 0.3021,
+      "step": 10100
+    },
+    {
+      "epoch": 63.75,
+      "learning_rate": 1.7509805797378744e-05,
+      "loss": 0.2972,
+      "step": 10200
+    },
+    {
+      "epoch": 64.0,
+      "eval_loss": 0.3553718328475952,
+      "eval_runtime": 135.8925,
+      "eval_samples_per_second": 11.038,
+      "eval_wer": 0.19806148853120478,
+      "step": 10240
+    },
+    {
+      "epoch": 64.37,
+      "learning_rate": 1.7366306323543482e-05,
+      "loss": 0.2987,
+      "step": 10300
+    },
+    {
+      "epoch": 65.0,
+      "learning_rate": 1.7222806849708216e-05,
+      "loss": 0.2951,
+      "step": 10400
+    },
+    {
+      "epoch": 65.0,
+      "eval_loss": 0.35284093022346497,
+      "eval_runtime": 136.2097,
+      "eval_samples_per_second": 11.012,
+      "eval_wer": 0.20051752116135257,
+      "step": 10400
+    },
+    {
+      "epoch": 65.62,
+      "learning_rate": 1.7079307375872954e-05,
+      "loss": 0.2972,
+      "step": 10500
+    },
+    {
+      "epoch": 66.0,
+      "eval_loss": 0.34928229451179504,
+      "eval_runtime": 135.0832,
+      "eval_samples_per_second": 11.104,
+      "eval_wer": 0.197842199903513,
+      "step": 10560
+    },
+    {
+      "epoch": 66.25,
+      "learning_rate": 1.6935807902037692e-05,
+      "loss": 0.2932,
+      "step": 10600
+    },
+    {
+      "epoch": 66.87,
+      "learning_rate": 1.679230842820243e-05,
+      "loss": 0.2923,
+      "step": 10700
+    },
+    {
+      "epoch": 67.0,
+      "eval_loss": 0.35474905371665955,
+      "eval_runtime": 136.3777,
+      "eval_samples_per_second": 10.999,
+      "eval_wer": 0.1951230209201351,
+      "step": 10720
+    },
+    {
+      "epoch": 67.5,
+      "learning_rate": 1.6648808954367167e-05,
+      "loss": 0.2944,
+      "step": 10800
+    },
+    {
+      "epoch": 68.0,
+      "eval_loss": 0.3355397582054138,
+      "eval_runtime": 136.8623,
+      "eval_samples_per_second": 10.96,
+      "eval_wer": 0.19354414280075435,
+      "step": 10880
+    },
+    {
+      "epoch": 68.12,
+      "learning_rate": 1.6505309480531905e-05,
+      "loss": 0.2921,
+      "step": 10900
+    },
+    {
+      "epoch": 68.75,
+      "learning_rate": 1.6361810006696643e-05,
+      "loss": 0.2898,
+      "step": 11000
+    },
+    {
+      "epoch": 69.0,
+      "eval_loss": 0.3218751549720764,
+      "eval_runtime": 134.7659,
+      "eval_samples_per_second": 11.13,
+      "eval_wer": 0.1917898337792202,
+      "step": 11040
+    },
+    {
+      "epoch": 69.37,
+      "learning_rate": 1.621831053286138e-05,
+      "loss": 0.2838,
+      "step": 11100
+    },
+    {
+      "epoch": 70.0,
+      "learning_rate": 1.607481105902612e-05,
+      "loss": 0.2841,
+      "step": 11200
+    },
+    {
+      "epoch": 70.0,
+      "eval_loss": 0.328848272562027,
+      "eval_runtime": 135.8246,
+      "eval_samples_per_second": 11.044,
+      "eval_wer": 0.1837638700057015,
+      "step": 11200
+    },
+    {
+      "epoch": 70.62,
+      "learning_rate": 1.5931311585190853e-05,
+      "loss": 0.2869,
+      "step": 11300
+    },
+    {
+      "epoch": 71.0,
+      "eval_loss": 0.3385452330112457,
+      "eval_runtime": 136.3804,
+      "eval_samples_per_second": 10.999,
+      "eval_wer": 0.19521073637121178,
+      "step": 11360
+    },
+    {
+      "epoch": 71.25,
+      "learning_rate": 1.578781211135559e-05,
+      "loss": 0.2824,
+      "step": 11400
+    },
+    {
+      "epoch": 71.87,
+      "learning_rate": 1.564431263752033e-05,
+      "loss": 0.2798,
+      "step": 11500
+    },
+    {
+      "epoch": 72.0,
+      "eval_loss": 0.33345770835876465,
+      "eval_runtime": 136.5311,
+      "eval_samples_per_second": 10.987,
+      "eval_wer": 0.1847287399675453,
+      "step": 11520
+    },
+    {
+      "epoch": 72.5,
+      "learning_rate": 1.5500813163685066e-05,
+      "loss": 0.279,
+      "step": 11600
+    },
+    {
+      "epoch": 73.0,
+      "eval_loss": 0.3369726836681366,
+      "eval_runtime": 136.9141,
+      "eval_samples_per_second": 10.956,
+      "eval_wer": 0.18871979299153546,
+      "step": 11680
+    },
+    {
+      "epoch": 73.12,
+      "learning_rate": 1.5357313689849804e-05,
+      "loss": 0.2778,
+      "step": 11700
+    },
+    {
+      "epoch": 73.75,
+      "learning_rate": 1.5213814216014542e-05,
+      "loss": 0.276,
+      "step": 11800
+    },
+    {
+      "epoch": 74.0,
+      "eval_loss": 0.33662334084510803,
+      "eval_runtime": 137.0118,
+      "eval_samples_per_second": 10.948,
+      "eval_wer": 0.19503530546905837,
+      "step": 11840
+    },
+    {
+      "epoch": 74.37,
+      "learning_rate": 1.507031474217928e-05,
+      "loss": 0.2792,
+      "step": 11900
+    },
+    {
+      "epoch": 75.0,
+      "learning_rate": 1.4926815268344016e-05,
+      "loss": 0.273,
+      "step": 12000
+    },
+    {
+      "epoch": 75.0,
+      "eval_loss": 0.3384232521057129,
+      "eval_runtime": 136.9469,
+      "eval_samples_per_second": 10.953,
+      "eval_wer": 0.18722863032323145,
+      "step": 12000
+    },
+    {
+      "epoch": 75.62,
+      "learning_rate": 1.4783315794508753e-05,
+      "loss": 0.2756,
+      "step": 12100
+    },
+    {
+      "epoch": 76.0,
+      "eval_loss": 0.33496129512786865,
+      "eval_runtime": 135.0526,
+      "eval_samples_per_second": 11.107,
+      "eval_wer": 0.18389544318231657,
+      "step": 12160
+    },
+    {
+      "epoch": 76.25,
+      "learning_rate": 1.4639816320673491e-05,
+      "loss": 0.2746,
+      "step": 12200
+    },
+    {
+      "epoch": 76.87,
+      "learning_rate": 1.4496316846838229e-05,
+      "loss": 0.2668,
+      "step": 12300
+    },
+    {
+      "epoch": 77.0,
+      "eval_loss": 0.3347358703613281,
+      "eval_runtime": 137.095,
+      "eval_samples_per_second": 10.941,
+      "eval_wer": 0.19091267926845315,
+      "step": 12320
+    },
+    {
+      "epoch": 77.5,
+      "learning_rate": 1.4352817373002967e-05,
+      "loss": 0.2713,
+      "step": 12400
+    },
+    {
+      "epoch": 78.0,
+      "eval_loss": 0.3463807702064514,
+      "eval_runtime": 135.9673,
+      "eval_samples_per_second": 11.032,
+      "eval_wer": 0.18714091487215473,
+      "step": 12480
+    },
+    {
+      "epoch": 78.12,
+      "learning_rate": 1.4209317899167704e-05,
+      "loss": 0.2636,
+      "step": 12500
+    },
+    {
+      "epoch": 78.75,
+      "learning_rate": 1.406581842533244e-05,
+      "loss": 0.2685,
+      "step": 12600
+    },
+    {
+      "epoch": 79.0,
+      "eval_loss": 0.35181865096092224,
+      "eval_runtime": 136.9541,
+      "eval_samples_per_second": 10.953,
+      "eval_wer": 0.19306170781983245,
+      "step": 12640
+    },
+    {
+      "epoch": 79.0,
+      "step": 12640,
+      "total_flos": 0,
+      "train_runtime": 205971.9249,
+      "train_samples_per_second": 0.109
+    }
+  ],
+  "max_steps": 22400,
+  "num_train_epochs": 140,
+  "total_flos": 0,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9592f65593e188d4a169b4c0656cde5e2706f06a580f398fe319347ae191788
+size 2543

vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "|": 1, "<unk>": 2, "a": 3, "b": 4, "c": 5, "d": 6, "e": 7, "f": 8, "g": 9, "h": 10, "i": 11, "j": 12, "k": 13, "l": 14, "m": 15, "n": 16, "o": 17, "p": 18, "q": 19, "r": 20, "s": 21, "t": 22, "u": 23, "v": 24, "w": 25, "x": 26, "y": 27, "z": 28, "ç": 29, "ã": 30, "à": 31, "á": 32, "â": 33, "ê": 34, "é": 35, "í": 36, "ó": 37, "ô": 38, "õ": 39, "ú": 40, "û": 41, "-": 42, "<s>": 43, "</s>": 44}