Add checkpoints

README.md

@@ -0,0 +1,69 @@
+---
+language: pt
+datasets:
+- Common Voice 
+metrics:
+- wer
+tags:
+- audio
+- speech
+- wav2vec2
+- pt
+- portuguese-speech-corpus
+- automatic-speech-recognition
+- speech
+- PyTorch
+license: apache-2.0
+model-index:
+- name: Edresson Casanova Wav2vec2 Large 100k Voxpopuli fine-tuned with a single-speaker dataset in Portuguese 
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    metrics:
+       - name: Test Common Voice 7.0 WER
+         type: wer
+         value: 63.90
+---
+
+# Wav2vec2 Large 100k Voxpopuli fine-tuned with a single-speaker dataset plus Data Augmentation in Portuguese 
+
+[Wav2vec2 Large 100k Voxpopuli](https://huggingface.co/facebook/wav2vec2-large-100k-voxpopuli) fine-tuned in Portuguese using a single-speaker dataset (TTS-Portuguese Corpus).
+
+
+
+# Use this model
+
+```python
+
+from transformers import AutoTokenizer, Wav2Vec2ForCTC
+  
+tokenizer = AutoTokenizer.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-portuguese")
+
+model = Wav2Vec2ForCTC.from_pretrained("Edresson/wav2vec2-large-100k-voxpopuli-ft-TTS-Dataset-portuguese")
+```
+# Results
+For the results check the [article (Soon)]()
+
+# Example test with Common Voice Dataset
+
+
+```python
+dataset = load_dataset("common_voice", "pt", split="test", data_dir="./cv-corpus-7.0-2021-07-21")
+
+resampler = torchaudio.transforms.Resample(orig_freq=48_000, new_freq=16_000)
+
+def map_to_array(batch):
+    speech, _ = torchaudio.load(batch["path"])
+    batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
+    batch["sampling_rate"] = resampler.new_freq
+    batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower().replace("’", "'")
+    return batch
+```
+
+```python
+ds = dataset.map(map_to_array)
+result = ds.map(map_to_pred, batched=True, batch_size=1, remove_columns=list(ds.features.keys()))
+print(wer.compute(predictions=result["predicted"], references=result["target"]))
+```
+

all_results.json

@@ -0,0 +1,23 @@
+{
+    "epoch": 129.99,
+    "eval_loss": 0.6900125741958618,
+    "eval_mem_cpu_alloc_delta": 152395776,
+    "eval_mem_cpu_peaked_delta": 122880,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 7244027392,
+    "eval_runtime": 49.6987,
+    "eval_samples": 500,
+    "eval_samples_per_second": 10.061,
+    "eval_wer": 0.4532554257095159,
+    "init_mem_cpu_alloc_delta": 3913703424,
+    "init_mem_cpu_peaked_delta": 1208139776,
+    "init_mem_gpu_alloc_delta": 1261939712,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 2117373952,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778624512,
+    "train_mem_gpu_peaked_delta": 9083709440,
+    "train_runtime": 94419.6153,
+    "train_samples": 3083,
+    "train_samples_per_second": 0.024
+}

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 +1,162 @@
+{
+    "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":"/raid/datasets/DA/musan/speech/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 13.0,
+            "max_snr_in_db": 20.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        {
+            "name": "additive",
+            "sounds_path":"/raid/datasets/DA/musan/music/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 5.0,
+            "max_snr_in_db": 15.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        {
+            "name": "additive",
+            "sounds_path":"/raid/datasets/DA/musan/noise/", // download: https://www.openslr.org/17/
+            "lru_cache_size": 32, // Maximum size of the LRU cache for storing noise files in memory
+            "min_snr_in_db": 0.0,
+            "max_snr_in_db": 15.0,
+            // "sample_rate": 16000,
+            "p": 0.25
+        },
+        // rir filter proposed by: https://ieeexplore.ieee.org/document/7953152
+        {
+            "name": "rir",
+            "ir_path": "/raid/datasets/DA/RIRS_NOISES/simulated_rirs/", // download: https://www.openslr.org/28/
+            "lru_cache_size": 128, // Maximum size of the LRU cache for storing noise files in memory
+            // "sample_rate": 16000,
+            "p": 0.25
+        }
+        , 
+        // {
+        //     "name": "gain",
+        //     "min_gain_in_db": -18.0,
+        //     "max_gain_in_db": 6,
+        //     "p": 0.25 // propability of apply this method, 0 is disable
+        // },
+        {
+            "name": "pitch_shift",
+            "min_semitones": -4,
+            "max_semitones": 4,
+            "p": 0.25 // propability of apply this method, 0 is disable
+        },
+        {
+            "name": "gaussian",
+            "min_amplitude": 0.0001,
+            "max_amplitude": 0.001,
+            "p": 0.25 // propability of apply this method, 0 is disable
+        }
+    ],
+    // PATHS
+    "output_path": "../checkpoints/YourTTS2ASR/Wav2Vec-voxpopuli/one-speaker/just-TTS/PT/140-epoch-high-bs/",
+    // CACHE
+    "dataset_cache": "../datasets/",
+
+    // DATASETS
+    "datasets":{ 
+        "files_path": "/raid/datasets/TTS-Portuguese-Corpus/", // relative path for audios It's will be join with the CS
+        "train":
+            [
+               // this dicts is pass directly for the load dataset see the documentation: https://huggingface.co/docs/datasets/package_reference/loading_methods.html#datasets.load_dataset 
+            {
+                "name": "csv",
+                "path": "csv",
+                
+                "data_files": ["/raid/datasets/TTS-Portuguese-Corpus/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": ["/raid/datasets/TTS-Portuguese-Corpus/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": ["/raid/datasets/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": 129.99,
+    "eval_loss": 0.6900125741958618,
+    "eval_mem_cpu_alloc_delta": 152395776,
+    "eval_mem_cpu_peaked_delta": 122880,
+    "eval_mem_gpu_alloc_delta": 0,
+    "eval_mem_gpu_peaked_delta": 7244027392,
+    "eval_runtime": 49.6987,
+    "eval_samples": 500,
+    "eval_samples_per_second": 10.061,
+    "eval_wer": 0.4532554257095159
+}

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 sha256:6a0b7787c4b812154e4d737af312cac8d7b1f3197d72e84feb6999aa3007be90
+size 1262108145

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": 129.99,
+    "init_mem_cpu_alloc_delta": 3913703424,
+    "init_mem_cpu_peaked_delta": 1208139776,
+    "init_mem_gpu_alloc_delta": 1261939712,
+    "init_mem_gpu_peaked_delta": 0,
+    "train_mem_cpu_alloc_delta": 2117373952,
+    "train_mem_cpu_peaked_delta": 24576,
+    "train_mem_gpu_alloc_delta": 3778624512,
+    "train_mem_gpu_peaked_delta": 9083709440,
+    "train_runtime": 94419.6153,
+    "train_samples": 3083,
+    "train_samples_per_second": 0.024
+}

trainer_state.json

@@ -0,0 +1,1189 @@
+{
+  "best_metric": 0.6735167503356934,
+  "best_model_checkpoint": "../checkpoints/YourTTS2ASR/Wav2Vec-voxpopuli/one-speaker/just-TTS/PT/140-epoch-high-bs/checkpoint-1920",
+  "epoch": 129.99481865284974,
+  "global_step": 2080,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.06,
+      "learning_rate": 2.0000000000000002e-07,
+      "loss": 12.0818,
+      "step": 1
+    },
+    {
+      "epoch": 0.99,
+      "eval_loss": 12.243678092956543,
+      "eval_runtime": 49.2931,
+      "eval_samples_per_second": 10.143,
+      "eval_wer": 1.0009390651085142,
+      "step": 16
+    },
+    {
+      "epoch": 1.99,
+      "eval_loss": 12.257490158081055,
+      "eval_runtime": 50.2087,
+      "eval_samples_per_second": 9.958,
+      "eval_wer": 1.0005217028380635,
+      "step": 32
+    },
+    {
+      "epoch": 2.99,
+      "eval_loss": 11.961856842041016,
+      "eval_runtime": 48.9816,
+      "eval_samples_per_second": 10.208,
+      "eval_wer": 0.9989565943238731,
+      "step": 48
+    },
+    {
+      "epoch": 3.99,
+      "eval_loss": 10.610284805297852,
+      "eval_runtime": 49.2335,
+      "eval_samples_per_second": 10.156,
+      "eval_wer": 0.9995826377295493,
+      "step": 64
+    },
+    {
+      "epoch": 4.99,
+      "eval_loss": 8.434402465820312,
+      "eval_runtime": 50.1138,
+      "eval_samples_per_second": 9.977,
+      "eval_wer": 1.0,
+      "step": 80
+    },
+    {
+      "epoch": 5.99,
+      "eval_loss": 6.819848537445068,
+      "eval_runtime": 50.2226,
+      "eval_samples_per_second": 9.956,
+      "eval_wer": 1.0,
+      "step": 96
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 11.0338,
+      "step": 100
+    },
+    {
+      "epoch": 6.99,
+      "eval_loss": 5.784036159515381,
+      "eval_runtime": 49.485,
+      "eval_samples_per_second": 10.104,
+      "eval_wer": 1.0,
+      "step": 112
+    },
+    {
+      "epoch": 7.99,
+      "eval_loss": 5.032914161682129,
+      "eval_runtime": 49.5792,
+      "eval_samples_per_second": 10.085,
+      "eval_wer": 1.0,
+      "step": 128
+    },
+    {
+      "epoch": 8.99,
+      "eval_loss": 4.575235366821289,
+      "eval_runtime": 48.3292,
+      "eval_samples_per_second": 10.346,
+      "eval_wer": 1.0,
+      "step": 144
+    },
+    {
+      "epoch": 9.99,
+      "eval_loss": 4.227362155914307,
+      "eval_runtime": 50.081,
+      "eval_samples_per_second": 9.984,
+      "eval_wer": 1.0,
+      "step": 160
+    },
+    {
+      "epoch": 10.99,
+      "eval_loss": 3.977447271347046,
+      "eval_runtime": 49.6321,
+      "eval_samples_per_second": 10.074,
+      "eval_wer": 1.0,
+      "step": 176
+    },
+    {
+      "epoch": 11.99,
+      "eval_loss": 3.8090567588806152,
+      "eval_runtime": 49.5094,
+      "eval_samples_per_second": 10.099,
+      "eval_wer": 1.0,
+      "step": 192
+    },
+    {
+      "epoch": 12.5,
+      "learning_rate": 2.9282296650717705e-05,
+      "loss": 5.2367,
+      "step": 200
+    },
+    {
+      "epoch": 12.99,
+      "eval_loss": 3.6537370681762695,
+      "eval_runtime": 49.2195,
+      "eval_samples_per_second": 10.159,
+      "eval_wer": 1.0,
+      "step": 208
+    },
+    {
+      "epoch": 13.99,
+      "eval_loss": 3.5310537815093994,
+      "eval_runtime": 48.9715,
+      "eval_samples_per_second": 10.21,
+      "eval_wer": 1.0,
+      "step": 224
+    },
+    {
+      "epoch": 14.99,
+      "eval_loss": 3.4466352462768555,
+      "eval_runtime": 49.4588,
+      "eval_samples_per_second": 10.109,
+      "eval_wer": 1.0,
+      "step": 240
+    },
+    {
+      "epoch": 15.99,
+      "eval_loss": 3.325173854827881,
+      "eval_runtime": 48.5474,
+      "eval_samples_per_second": 10.299,
+      "eval_wer": 1.0,
+      "step": 256
+    },
+    {
+      "epoch": 16.99,
+      "eval_loss": 3.2516260147094727,
+      "eval_runtime": 49.5175,
+      "eval_samples_per_second": 10.097,
+      "eval_wer": 1.0,
+      "step": 272
+    },
+    {
+      "epoch": 17.99,
+      "eval_loss": 3.1839869022369385,
+      "eval_runtime": 48.9663,
+      "eval_samples_per_second": 10.211,
+      "eval_wer": 1.0,
+      "step": 288
+    },
+    {
+      "epoch": 18.75,
+      "learning_rate": 2.784688995215311e-05,
+      "loss": 3.5824,
+      "step": 300
+    },
+    {
+      "epoch": 18.99,
+      "eval_loss": 3.1358871459960938,
+      "eval_runtime": 49.4599,
+      "eval_samples_per_second": 10.109,
+      "eval_wer": 1.0,
+      "step": 304
+    },
+    {
+      "epoch": 19.99,
+      "eval_loss": 3.0906283855438232,
+      "eval_runtime": 49.5292,
+      "eval_samples_per_second": 10.095,
+      "eval_wer": 1.0,
+      "step": 320
+    },
+    {
+      "epoch": 20.99,
+      "eval_loss": 3.0470166206359863,
+      "eval_runtime": 49.8959,
+      "eval_samples_per_second": 10.021,
+      "eval_wer": 1.0,
+      "step": 336
+    },
+    {
+      "epoch": 21.99,
+      "eval_loss": 3.0199356079101562,
+      "eval_runtime": 48.7586,
+      "eval_samples_per_second": 10.255,
+      "eval_wer": 1.0,
+      "step": 352
+    },
+    {
+      "epoch": 22.99,
+      "eval_loss": 2.993663787841797,
+      "eval_runtime": 49.2925,
+      "eval_samples_per_second": 10.144,
+      "eval_wer": 1.0,
+      "step": 368
+    },
+    {
+      "epoch": 23.99,
+      "eval_loss": 2.9856507778167725,
+      "eval_runtime": 48.889,
+      "eval_samples_per_second": 10.227,
+      "eval_wer": 1.0,
+      "step": 384
+    },
+    {
+      "epoch": 24.99,
+      "learning_rate": 2.6411483253588518e-05,
+      "loss": 3.088,
+      "step": 400
+    },
+    {
+      "epoch": 24.99,
+      "eval_loss": 2.952263116836548,
+      "eval_runtime": 49.1389,
+      "eval_samples_per_second": 10.175,
+      "eval_wer": 1.0,
+      "step": 400
+    },
+    {
+      "epoch": 25.99,
+      "eval_loss": 2.9487361907958984,
+      "eval_runtime": 49.3941,
+      "eval_samples_per_second": 10.123,
+      "eval_wer": 1.0,
+      "step": 416
+    },
+    {
+      "epoch": 26.99,
+      "eval_loss": 2.9333157539367676,
+      "eval_runtime": 49.2595,
+      "eval_samples_per_second": 10.15,
+      "eval_wer": 1.0,
+      "step": 432
+    },
+    {
+      "epoch": 27.99,
+      "eval_loss": 2.9365267753601074,
+      "eval_runtime": 49.5918,
+      "eval_samples_per_second": 10.082,
+      "eval_wer": 1.0,
+      "step": 448
+    },
+    {
+      "epoch": 28.99,
+      "eval_loss": 2.926832914352417,
+      "eval_runtime": 49.5193,
+      "eval_samples_per_second": 10.097,
+      "eval_wer": 1.0,
+      "step": 464
+    },
+    {
+      "epoch": 29.99,
+      "eval_loss": 2.9162416458129883,
+      "eval_runtime": 49.57,
+      "eval_samples_per_second": 10.087,
+      "eval_wer": 1.0,
+      "step": 480
+    },
+    {
+      "epoch": 30.99,
+      "eval_loss": 2.9036505222320557,
+      "eval_runtime": 49.6143,
+      "eval_samples_per_second": 10.078,
+      "eval_wer": 1.0,
+      "step": 496
+    },
+    {
+      "epoch": 31.25,
+      "learning_rate": 2.4976076555023923e-05,
+      "loss": 2.9504,
+      "step": 500
+    },
+    {
+      "epoch": 31.99,
+      "eval_loss": 2.9000539779663086,
+      "eval_runtime": 49.4941,
+      "eval_samples_per_second": 10.102,
+      "eval_wer": 1.0,
+      "step": 512
+    },
+    {
+      "epoch": 32.99,
+      "eval_loss": 2.893620491027832,
+      "eval_runtime": 50.1869,
+      "eval_samples_per_second": 9.963,
+      "eval_wer": 1.0,
+      "step": 528
+    },
+    {
+      "epoch": 33.99,
+      "eval_loss": 2.8881995677948,
+      "eval_runtime": 49.8317,
+      "eval_samples_per_second": 10.034,
+      "eval_wer": 1.0,
+      "step": 544
+    },
+    {
+      "epoch": 34.99,
+      "eval_loss": 2.8850040435791016,
+      "eval_runtime": 49.2167,
+      "eval_samples_per_second": 10.159,
+      "eval_wer": 1.0,
+      "step": 560
+    },
+    {
+      "epoch": 35.99,
+      "eval_loss": 2.8831725120544434,
+      "eval_runtime": 49.9233,
+      "eval_samples_per_second": 10.015,
+      "eval_wer": 1.0,
+      "step": 576
+    },
+    {
+      "epoch": 36.99,
+      "eval_loss": 2.8862743377685547,
+      "eval_runtime": 49.9603,
+      "eval_samples_per_second": 10.008,
+      "eval_wer": 1.0,
+      "step": 592
+    },
+    {
+      "epoch": 37.5,
+      "learning_rate": 2.354066985645933e-05,
+      "loss": 2.9116,
+      "step": 600
+    },
+    {
+      "epoch": 37.99,
+      "eval_loss": 2.8713128566741943,
+      "eval_runtime": 48.8267,
+      "eval_samples_per_second": 10.24,
+      "eval_wer": 1.0,
+      "step": 608
+    },
+    {
+      "epoch": 38.99,
+      "eval_loss": 2.8694326877593994,
+      "eval_runtime": 50.3391,
+      "eval_samples_per_second": 9.933,
+      "eval_wer": 1.0,
+      "step": 624
+    },
+    {
+      "epoch": 39.99,
+      "eval_loss": 2.8684744834899902,
+      "eval_runtime": 49.7451,
+      "eval_samples_per_second": 10.051,
+      "eval_wer": 1.0,
+      "step": 640
+    },
+    {
+      "epoch": 40.99,
+      "eval_loss": 2.8572192192077637,
+      "eval_runtime": 50.1709,
+      "eval_samples_per_second": 9.966,
+      "eval_wer": 1.0,
+      "step": 656
+    },
+    {
+      "epoch": 41.99,
+      "eval_loss": 2.8502085208892822,
+      "eval_runtime": 49.5484,
+      "eval_samples_per_second": 10.091,
+      "eval_wer": 1.0,
+      "step": 672
+    },
+    {
+      "epoch": 42.99,
+      "eval_loss": 2.8417484760284424,
+      "eval_runtime": 49.0891,
+      "eval_samples_per_second": 10.186,
+      "eval_wer": 1.0,
+      "step": 688
+    },
+    {
+      "epoch": 43.75,
+      "learning_rate": 2.2105263157894736e-05,
+      "loss": 2.8836,
+      "step": 700
+    },
+    {
+      "epoch": 43.99,
+      "eval_loss": 2.8355417251586914,
+      "eval_runtime": 49.7,
+      "eval_samples_per_second": 10.06,
+      "eval_wer": 1.0,
+      "step": 704
+    },
+    {
+      "epoch": 44.99,
+      "eval_loss": 2.8320789337158203,
+      "eval_runtime": 49.7957,
+      "eval_samples_per_second": 10.041,
+      "eval_wer": 1.0,
+      "step": 720
+    },
+    {
+      "epoch": 45.99,
+      "eval_loss": 2.8173515796661377,
+      "eval_runtime": 49.7947,
+      "eval_samples_per_second": 10.041,
+      "eval_wer": 1.0,
+      "step": 736
+    },
+    {
+      "epoch": 46.99,
+      "eval_loss": 2.799645185470581,
+      "eval_runtime": 49.4675,
+      "eval_samples_per_second": 10.108,
+      "eval_wer": 1.0,
+      "step": 752
+    },
+    {
+      "epoch": 47.99,
+      "eval_loss": 2.783804178237915,
+      "eval_runtime": 49.2068,
+      "eval_samples_per_second": 10.161,
+      "eval_wer": 1.0,
+      "step": 768
+    },
+    {
+      "epoch": 48.99,
+      "eval_loss": 2.759208917617798,
+      "eval_runtime": 49.4097,
+      "eval_samples_per_second": 10.119,
+      "eval_wer": 1.0,
+      "step": 784
+    },
+    {
+      "epoch": 49.99,
+      "learning_rate": 2.0669856459330144e-05,
+      "loss": 2.8277,
+      "step": 800
+    },
+    {
+      "epoch": 49.99,
+      "eval_loss": 2.7332029342651367,
+      "eval_runtime": 49.4948,
+      "eval_samples_per_second": 10.102,
+      "eval_wer": 1.0,
+      "step": 800
+    },
+    {
+      "epoch": 50.99,
+      "eval_loss": 2.709989309310913,
+      "eval_runtime": 48.8084,
+      "eval_samples_per_second": 10.244,
+      "eval_wer": 0.9984348914858097,
+      "step": 816
+    },
+    {
+      "epoch": 51.99,
+      "eval_loss": 2.6740872859954834,
+      "eval_runtime": 49.8376,
+      "eval_samples_per_second": 10.033,
+      "eval_wer": 0.9957220367278798,
+      "step": 832
+    },
+    {
+      "epoch": 52.99,
+      "eval_loss": 2.646063804626465,
+      "eval_runtime": 49.3421,
+      "eval_samples_per_second": 10.133,
+      "eval_wer": 0.9908180300500835,
+      "step": 848
+    },
+    {
+      "epoch": 53.99,
+      "eval_loss": 2.6031899452209473,
+      "eval_runtime": 49.5023,
+      "eval_samples_per_second": 10.101,
+      "eval_wer": 0.9760016694490818,
+      "step": 864
+    },
+    {
+      "epoch": 54.99,
+      "eval_loss": 2.5573575496673584,
+      "eval_runtime": 49.9478,
+      "eval_samples_per_second": 10.01,
+      "eval_wer": 0.9698455759599333,
+      "step": 880
+    },
+    {
+      "epoch": 55.99,
+      "eval_loss": 2.488868236541748,
+      "eval_runtime": 50.1085,
+      "eval_samples_per_second": 9.978,
+      "eval_wer": 0.9582637729549248,
+      "step": 896
+    },
+    {
+      "epoch": 56.25,
+      "learning_rate": 1.9234449760765553e-05,
+      "loss": 2.6844,
+      "step": 900
+    },
+    {
+      "epoch": 56.99,
+      "eval_loss": 2.4326117038726807,
+      "eval_runtime": 50.1267,
+      "eval_samples_per_second": 9.975,
+      "eval_wer": 0.9516903171953256,
+      "step": 912
+    },
+    {
+      "epoch": 57.99,
+      "eval_loss": 2.357138156890869,
+      "eval_runtime": 50.0431,
+      "eval_samples_per_second": 9.991,
+      "eval_wer": 0.9559682804674458,
+      "step": 928
+    },
+    {
+      "epoch": 58.99,
+      "eval_loss": 2.2913711071014404,
+      "eval_runtime": 50.0784,
+      "eval_samples_per_second": 9.984,
+      "eval_wer": 0.9559682804674458,
+      "step": 944
+    },
+    {
+      "epoch": 59.99,
+      "eval_loss": 2.21211314201355,
+      "eval_runtime": 49.4565,
+      "eval_samples_per_second": 10.11,
+      "eval_wer": 0.9565943238731218,
+      "step": 960
+    },
+    {
+      "epoch": 60.99,
+      "eval_loss": 2.1155571937561035,
+      "eval_runtime": 48.9205,
+      "eval_samples_per_second": 10.221,
+      "eval_wer": 0.9346828046744574,
+      "step": 976
+    },
+    {
+      "epoch": 61.99,
+      "eval_loss": 2.0269429683685303,
+      "eval_runtime": 49.2233,
+      "eval_samples_per_second": 10.158,
+      "eval_wer": 0.9186143572621035,
+      "step": 992
+    },
+    {
+      "epoch": 62.5,
+      "learning_rate": 1.7799043062200958e-05,
+      "loss": 2.3941,
+      "step": 1000
+    },
+    {
+      "epoch": 62.99,
+      "eval_loss": 1.9515246152877808,
+      "eval_runtime": 49.3099,
+      "eval_samples_per_second": 10.14,
+      "eval_wer": 0.9009808013355592,
+      "step": 1008
+    },
+    {
+      "epoch": 63.99,
+      "eval_loss": 1.8633095026016235,
+      "eval_runtime": 49.6022,
+      "eval_samples_per_second": 10.08,
+      "eval_wer": 0.8821994991652755,
+      "step": 1024
+    },
+    {
+      "epoch": 64.99,
+      "eval_loss": 1.7742440700531006,
+      "eval_runtime": 49.4148,
+      "eval_samples_per_second": 10.118,
+      "eval_wer": 0.8746869782971619,
+      "step": 1040
+    },
+    {
+      "epoch": 65.99,
+      "eval_loss": 1.6952035427093506,
+      "eval_runtime": 49.4954,
+      "eval_samples_per_second": 10.102,
+      "eval_wer": 0.8543405676126878,
+      "step": 1056
+    },
+    {
+      "epoch": 66.99,
+      "eval_loss": 1.6496139764785767,
+      "eval_runtime": 50.2365,
+      "eval_samples_per_second": 9.953,
+      "eval_wer": 0.8415066777963273,
+      "step": 1072
+    },
+    {
+      "epoch": 67.99,
+      "eval_loss": 1.577644944190979,
+      "eval_runtime": 49.0048,
+      "eval_samples_per_second": 10.203,
+      "eval_wer": 0.8229340567612687,
+      "step": 1088
+    },
+    {
+      "epoch": 68.75,
+      "learning_rate": 1.6363636363636363e-05,
+      "loss": 1.9737,
+      "step": 1100
+    },
+    {
+      "epoch": 68.99,
+      "eval_loss": 1.5079734325408936,
+      "eval_runtime": 49.8276,
+      "eval_samples_per_second": 10.035,
+      "eval_wer": 0.7996661101836394,
+      "step": 1104
+    },
+    {
+      "epoch": 69.99,
+      "eval_loss": 1.4567533731460571,
+      "eval_runtime": 48.81,
+      "eval_samples_per_second": 10.244,
+      "eval_wer": 0.7835976627712855,
+      "step": 1120
+    },
+    {
+      "epoch": 70.99,
+      "eval_loss": 1.4095492362976074,
+      "eval_runtime": 50.121,
+      "eval_samples_per_second": 9.976,
+      "eval_wer": 0.7838063439065108,
+      "step": 1136
+    },
+    {
+      "epoch": 71.99,
+      "eval_loss": 1.3987743854522705,
+      "eval_runtime": 49.582,
+      "eval_samples_per_second": 10.084,
+      "eval_wer": 0.7634599332220368,
+      "step": 1152
+    },
+    {
+      "epoch": 72.99,
+      "eval_loss": 1.3577879667282104,
+      "eval_runtime": 49.6213,
+      "eval_samples_per_second": 10.076,
+      "eval_wer": 0.7589732888146912,
+      "step": 1168
+    },
+    {
+      "epoch": 73.99,
+      "eval_loss": 1.3037357330322266,
+      "eval_runtime": 49.3802,
+      "eval_samples_per_second": 10.126,
+      "eval_wer": 0.7461393989983306,
+      "step": 1184
+    },
+    {
+      "epoch": 74.99,
+      "learning_rate": 1.492822966507177e-05,
+      "loss": 1.6567,
+      "step": 1200
+    },
+    {
+      "epoch": 74.99,
+      "eval_loss": 1.2708954811096191,
+      "eval_runtime": 49.4247,
+      "eval_samples_per_second": 10.116,
+      "eval_wer": 0.7350792988313857,
+      "step": 1200
+    },
+    {
+      "epoch": 75.99,
+      "eval_loss": 1.2239230871200562,
+      "eval_runtime": 50.0296,
+      "eval_samples_per_second": 9.994,
+      "eval_wer": 0.7258973288814691,
+      "step": 1216
+    },
+    {
+      "epoch": 76.99,
+      "eval_loss": 1.1831496953964233,
+      "eval_runtime": 49.5093,
+      "eval_samples_per_second": 10.099,
+      "eval_wer": 0.7141068447412354,
+      "step": 1232
+    },
+    {
+      "epoch": 77.99,
+      "eval_loss": 1.1715855598449707,
+      "eval_runtime": 50.4813,
+      "eval_samples_per_second": 9.905,
+      "eval_wer": 0.7060726210350584,
+      "step": 1248
+    },
+    {
+      "epoch": 78.99,
+      "eval_loss": 1.1489920616149902,
+      "eval_runtime": 49.5198,
+      "eval_samples_per_second": 10.097,
+      "eval_wer": 0.7013772954924875,
+      "step": 1264
+    },
+    {
+      "epoch": 79.99,
+      "eval_loss": 1.1155155897140503,
+      "eval_runtime": 49.0369,
+      "eval_samples_per_second": 10.196,
+      "eval_wer": 0.6840567612687813,
+      "step": 1280
+    },
+    {
+      "epoch": 80.99,
+      "eval_loss": 1.091223955154419,
+      "eval_runtime": 49.9086,
+      "eval_samples_per_second": 10.018,
+      "eval_wer": 0.6698664440734557,
+      "step": 1296
+    },
+    {
+      "epoch": 81.25,
+      "learning_rate": 1.3492822966507177e-05,
+      "loss": 1.4388,
+      "step": 1300
+    },
+    {
+      "epoch": 81.99,
+      "eval_loss": 1.0740453004837036,
+      "eval_runtime": 49.6299,
+      "eval_samples_per_second": 10.075,
+      "eval_wer": 0.659432387312187,
+      "step": 1312
+    },
+    {
+      "epoch": 82.99,
+      "eval_loss": 1.0269348621368408,
+      "eval_runtime": 49.8552,
+      "eval_samples_per_second": 10.029,
+      "eval_wer": 0.648059265442404,
+      "step": 1328
+    },
+    {
+      "epoch": 83.99,
+      "eval_loss": 0.9962567090988159,
+      "eval_runtime": 49.19,
+      "eval_samples_per_second": 10.165,
+      "eval_wer": 0.6295909849749582,
+      "step": 1344
+    },
+    {
+      "epoch": 84.99,
+      "eval_loss": 0.989078164100647,
+      "eval_runtime": 49.4835,
+      "eval_samples_per_second": 10.104,
+      "eval_wer": 0.6127921535893155,
+      "step": 1360
+    },
+    {
+      "epoch": 85.99,
+      "eval_loss": 0.9740233421325684,
+      "eval_runtime": 49.61,
+      "eval_samples_per_second": 10.079,
+      "eval_wer": 0.6173831385642737,
+      "step": 1376
+    },
+    {
+      "epoch": 86.99,
+      "eval_loss": 0.9527219533920288,
+      "eval_runtime": 49.7235,
+      "eval_samples_per_second": 10.056,
+      "eval_wer": 0.6007929883138564,
+      "step": 1392
+    },
+    {
+      "epoch": 87.5,
+      "learning_rate": 1.2057416267942584e-05,
+      "loss": 1.2741,
+      "step": 1400
+    },
+    {
+      "epoch": 87.99,
+      "eval_loss": 0.926001787185669,
+      "eval_runtime": 50.147,
+      "eval_samples_per_second": 9.971,
+      "eval_wer": 0.5874373956594324,
+      "step": 1408
+    },
+    {
+      "epoch": 88.99,
+      "eval_loss": 0.928646445274353,
+      "eval_runtime": 50.0432,
+      "eval_samples_per_second": 9.991,
+      "eval_wer": 0.5846202003338898,
+      "step": 1424
+    },
+    {
+      "epoch": 89.99,
+      "eval_loss": 0.915071964263916,
+      "eval_runtime": 48.6625,
+      "eval_samples_per_second": 10.275,
+      "eval_wer": 0.5746035058430717,
+      "step": 1440
+    },
+    {
+      "epoch": 90.99,
+      "eval_loss": 0.8867021203041077,
+      "eval_runtime": 49.6538,
+      "eval_samples_per_second": 10.07,
+      "eval_wer": 0.5595784641068448,
+      "step": 1456
+    },
+    {
+      "epoch": 91.99,
+      "eval_loss": 0.8880752921104431,
+      "eval_runtime": 49.9668,
+      "eval_samples_per_second": 10.007,
+      "eval_wer": 0.5575959933222037,
+      "step": 1472
+    },
+    {
+      "epoch": 92.99,
+      "eval_loss": 0.8759620189666748,
+      "eval_runtime": 49.4232,
+      "eval_samples_per_second": 10.117,
+      "eval_wer": 0.550813856427379,
+      "step": 1488
+    },
+    {
+      "epoch": 93.75,
+      "learning_rate": 1.062200956937799e-05,
+      "loss": 1.1621,
+      "step": 1500
+    },
+    {
+      "epoch": 93.99,
+      "eval_loss": 0.8622854948043823,
+      "eval_runtime": 50.0666,
+      "eval_samples_per_second": 9.987,
+      "eval_wer": 0.5416318864774624,
+      "step": 1504
+    },
+    {
+      "epoch": 94.99,
+      "eval_loss": 0.829269528388977,
+      "eval_runtime": 49.9208,
+      "eval_samples_per_second": 10.016,
+      "eval_wer": 0.5364148580968281,
+      "step": 1520
+    },
+    {
+      "epoch": 95.99,
+      "eval_loss": 0.8396909832954407,
+      "eval_runtime": 48.9453,
+      "eval_samples_per_second": 10.215,
+      "eval_wer": 0.5285893155258765,
+      "step": 1536
+    },
+    {
+      "epoch": 96.99,
+      "eval_loss": 0.8327041268348694,
+      "eval_runtime": 49.8765,
+      "eval_samples_per_second": 10.025,
+      "eval_wer": 0.5255634390651085,
+      "step": 1552
+    },
+    {
+      "epoch": 97.99,
+      "eval_loss": 0.8415578603744507,
+      "eval_runtime": 49.2064,
+      "eval_samples_per_second": 10.161,
+      "eval_wer": 0.528067612687813,
+      "step": 1568
+    },
+    {
+      "epoch": 98.99,
+      "eval_loss": 0.8029292225837708,
+      "eval_runtime": 49.3718,
+      "eval_samples_per_second": 10.127,
+      "eval_wer": 0.5109557595993323,
+      "step": 1584
+    },
+    {
+      "epoch": 99.99,
+      "learning_rate": 9.186602870813397e-06,
+      "loss": 1.0779,
+      "step": 1600
+    },
+    {
+      "epoch": 99.99,
+      "eval_loss": 0.8305151462554932,
+      "eval_runtime": 50.2704,
+      "eval_samples_per_second": 9.946,
+      "eval_wer": 0.5202420701168614,
+      "step": 1600
+    },
+    {
+      "epoch": 100.99,
+      "eval_loss": 0.794418215751648,
+      "eval_runtime": 48.8334,
+      "eval_samples_per_second": 10.239,
+      "eval_wer": 0.5020868113522537,
+      "step": 1616
+    },
+    {
+      "epoch": 101.99,
+      "eval_loss": 0.8060072064399719,
+      "eval_runtime": 50.0476,
+      "eval_samples_per_second": 9.99,
+      "eval_wer": 0.5134599332220368,
+      "step": 1632
+    },
+    {
+      "epoch": 102.99,
+      "eval_loss": 0.7654790878295898,
+      "eval_runtime": 49.2057,
+      "eval_samples_per_second": 10.161,
+      "eval_wer": 0.4867487479131887,
+      "step": 1648
+    },
+    {
+      "epoch": 103.99,
+      "eval_loss": 0.7860442996025085,
+      "eval_runtime": 49.9368,
+      "eval_samples_per_second": 10.013,
+      "eval_wer": 0.49947829716193654,
+      "step": 1664
+    },
+    {
+      "epoch": 104.99,
+      "eval_loss": 0.7663349509239197,
+      "eval_runtime": 49.0409,
+      "eval_samples_per_second": 10.196,
+      "eval_wer": 0.4897746243739566,
+      "step": 1680
+    },
+    {
+      "epoch": 105.99,
+      "eval_loss": 0.7740986347198486,
+      "eval_runtime": 49.5176,
+      "eval_samples_per_second": 10.097,
+      "eval_wer": 0.4820534223706177,
+      "step": 1696
+    },
+    {
+      "epoch": 106.25,
+      "learning_rate": 7.751196172248804e-06,
+      "loss": 1.0149,
+      "step": 1700
+    },
+    {
+      "epoch": 106.99,
+      "eval_loss": 0.72774338722229,
+      "eval_runtime": 49.0666,
+      "eval_samples_per_second": 10.19,
+      "eval_wer": 0.4803839732888147,
+      "step": 1712
+    },
+    {
+      "epoch": 107.99,
+      "eval_loss": 0.7349050641059875,
+      "eval_runtime": 49.3653,
+      "eval_samples_per_second": 10.129,
+      "eval_wer": 0.46869782971619367,
+      "step": 1728
+    },
+    {
+      "epoch": 108.99,
+      "eval_loss": 0.730620265007019,
+      "eval_runtime": 48.8671,
+      "eval_samples_per_second": 10.232,
+      "eval_wer": 0.4691151919866444,
+      "step": 1744
+    },
+    {
+      "epoch": 109.99,
+      "eval_loss": 0.7221301198005676,
+      "eval_runtime": 50.0918,
+      "eval_samples_per_second": 9.982,
+      "eval_wer": 0.4692195325542571,
+      "step": 1760
+    },
+    {
+      "epoch": 110.99,
+      "eval_loss": 0.7237880825996399,
+      "eval_runtime": 49.0461,
+      "eval_samples_per_second": 10.194,
+      "eval_wer": 0.4645242070116861,
+      "step": 1776
+    },
+    {
+      "epoch": 111.99,
+      "eval_loss": 0.7305303812026978,
+      "eval_runtime": 49.3111,
+      "eval_samples_per_second": 10.14,
+      "eval_wer": 0.46128964941569284,
+      "step": 1792
+    },
+    {
+      "epoch": 112.5,
+      "learning_rate": 6.31578947368421e-06,
+      "loss": 0.9691,
+      "step": 1800
+    },
+    {
+      "epoch": 112.99,
+      "eval_loss": 0.7333260774612427,
+      "eval_runtime": 49.8629,
+      "eval_samples_per_second": 10.027,
+      "eval_wer": 0.46295909849749584,
+      "step": 1808
+    },
+    {
+      "epoch": 113.99,
+      "eval_loss": 0.738413393497467,
+      "eval_runtime": 49.8063,
+      "eval_samples_per_second": 10.039,
+      "eval_wer": 0.4565943238731219,
+      "step": 1824
+    },
+    {
+      "epoch": 114.99,
+      "eval_loss": 0.7188318371772766,
+      "eval_runtime": 49.6388,
+      "eval_samples_per_second": 10.073,
+      "eval_wer": 0.46212437395659434,
+      "step": 1840
+    },
+    {
+      "epoch": 115.99,
+      "eval_loss": 0.6936877369880676,
+      "eval_runtime": 50.2237,
+      "eval_samples_per_second": 9.955,
+      "eval_wer": 0.4554465776293823,
+      "step": 1856
+    },
+    {
+      "epoch": 116.99,
+      "eval_loss": 0.7130375504493713,
+      "eval_runtime": 49.1949,
+      "eval_samples_per_second": 10.164,
+      "eval_wer": 0.4492904841402337,
+      "step": 1872
+    },
+    {
+      "epoch": 117.99,
+      "eval_loss": 0.7151590585708618,
+      "eval_runtime": 49.9104,
+      "eval_samples_per_second": 10.018,
+      "eval_wer": 0.4492904841402337,
+      "step": 1888
+    },
+    {
+      "epoch": 118.75,
+      "learning_rate": 4.880382775119617e-06,
+      "loss": 0.9391,
+      "step": 1900
+    },
+    {
+      "epoch": 118.99,
+      "eval_loss": 0.7040448188781738,
+      "eval_runtime": 49.7002,
+      "eval_samples_per_second": 10.06,
+      "eval_wer": 0.45398580968280466,
+      "step": 1904
+    },
+    {
+      "epoch": 119.99,
+      "eval_loss": 0.6735167503356934,
+      "eval_runtime": 49.7634,
+      "eval_samples_per_second": 10.048,
+      "eval_wer": 0.4393781302170284,
+      "step": 1920
+    },
+    {
+      "epoch": 120.99,
+      "eval_loss": 0.6741155385971069,
+      "eval_runtime": 49.3263,
+      "eval_samples_per_second": 10.137,
+      "eval_wer": 0.4453255425709516,
+      "step": 1936
+    },
+    {
+      "epoch": 121.99,
+      "eval_loss": 0.7235249280929565,
+      "eval_runtime": 49.3356,
+      "eval_samples_per_second": 10.135,
+      "eval_wer": 0.4561769616026711,
+      "step": 1952
+    },
+    {
+      "epoch": 122.99,
+      "eval_loss": 0.6766911745071411,
+      "eval_runtime": 49.3909,
+      "eval_samples_per_second": 10.123,
+      "eval_wer": 0.43541318864774625,
+      "step": 1968
+    },
+    {
+      "epoch": 123.99,
+      "eval_loss": 0.6751864552497864,
+      "eval_runtime": 49.3541,
+      "eval_samples_per_second": 10.131,
+      "eval_wer": 0.43343071786310516,
+      "step": 1984
+    },
+    {
+      "epoch": 124.99,
+      "learning_rate": 3.444976076555024e-06,
+      "loss": 0.9094,
+      "step": 2000
+    },
+    {
+      "epoch": 124.99,
+      "eval_loss": 0.6876149773597717,
+      "eval_runtime": 49.706,
+      "eval_samples_per_second": 10.059,
+      "eval_wer": 0.4397954924874791,
+      "step": 2000
+    },
+    {
+      "epoch": 125.99,
+      "eval_loss": 0.7010783553123474,
+      "eval_runtime": 49.0784,
+      "eval_samples_per_second": 10.188,
+      "eval_wer": 0.44835141903171954,
+      "step": 2016
+    },
+    {
+      "epoch": 126.99,
+      "eval_loss": 0.6755461692810059,
+      "eval_runtime": 48.3236,
+      "eval_samples_per_second": 10.347,
+      "eval_wer": 0.42497913188647746,
+      "step": 2032
+    },
+    {
+      "epoch": 127.99,
+      "eval_loss": 0.7173025608062744,
+      "eval_runtime": 49.6709,
+      "eval_samples_per_second": 10.066,
+      "eval_wer": 0.4476210350584307,
+      "step": 2048
+    },
+    {
+      "epoch": 128.99,
+      "eval_loss": 0.7063612937927246,
+      "eval_runtime": 49.982,
+      "eval_samples_per_second": 10.004,
+      "eval_wer": 0.44209098497495825,
+      "step": 2064
+    },
+    {
+      "epoch": 129.99,
+      "eval_loss": 0.6794615387916565,
+      "eval_runtime": 49.1686,
+      "eval_samples_per_second": 10.169,
+      "eval_wer": 0.43426544240400666,
+      "step": 2080
+    },
+    {
+      "epoch": 129.99,
+      "step": 2080,
+      "total_flos": 0,
+      "train_runtime": 94419.6153,
+      "train_samples_per_second": 0.024
+    }
+  ],
+  "max_steps": 2240,
+  "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:c1c71efb6adb93df87f465dc27c83d6ccf6bf7c561771e4f59ccb252db4cd828
+size 2607

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}