add wav2vec asr without language model

README.md

@@ -0,0 +1,93 @@
+---
+license: apache-2.0
+tags:
+- google/fleurs
+- generated_from_trainer
+- automatic-speech-recognition
+- hf-asr-leaderboard
+- pashto
+- ps
+datasets:
+- fleurs
+metrics:
+- wer
+model-index:
+- name: facebook/wav2vec2-xls-r-300m
+  results:
+  - task:
+      name: Automatic Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: google/fleurs
+      type: google/fleurs
+      args: 'config: ps_af, split: test'
+    metrics:
+    - name: Wer
+      type: wer
+      value: 0.5159447476125512
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# facebook/wav2vec2-xls-r-300m
+
+This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the GOOGLE/FLEURS - PS_AF dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.9162
+- Wer: 0.5159
+- Cer: 0.1972
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 7.5e-07
+- train_batch_size: 16
+- eval_batch_size: 16
+- seed: 42
+- gradient_accumulation_steps: 2
+- total_train_batch_size: 32
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- lr_scheduler_warmup_steps: 1000
+- training_steps: 6000
+- mixed_precision_training: Native AMP
+
+### Training results
+
+| Training Loss | Epoch | Step | Cer    | Validation Loss | Wer    |
+|:-------------:|:-----:|:----:|:------:|:---------------:|:------:|
+| 5.0767        | 6.33  | 500  | 1.0    | 4.8783          | 1.0    |
+| 3.1156        | 12.66 | 1000 | 1.0    | 3.0990          | 1.0    |
+| 1.3506        | 18.99 | 1500 | 0.2889 | 1.1056          | 0.7031 |
+| 0.9997        | 25.32 | 2000 | 0.2301 | 0.9191          | 0.5944 |
+| 0.7838        | 31.65 | 2500 | 0.2152 | 0.8952          | 0.5556 |
+| 0.6665        | 37.97 | 3000 | 0.2017 | 0.8908          | 0.5252 |
+| 0.6265        | 44.3  | 3500 | 0.1954 | 0.9063          | 0.5133 |
+| 0.5935        | 50.63 | 4000 | 0.1969 | 0.9162          | 0.5156 |
+| 0.5174        | 56.96 | 4500 | 0.1972 | 0.9287          | 0.5140 |
+| 0.5462        | 63.29 | 5000 | 0.1974 | 0.9370          | 0.5138 |
+| 0.5564        | 69.62 | 5500 | 0.1977 | 0.9461          | 0.5148 |
+| 0.5252        | 75.95 | 6000 | 0.9505 | 0.5118          | 0.1969 |
+
+
+### Framework versions
+
+- Transformers 4.26.0.dev0
+- Pytorch 1.13.1+cu117
+- Datasets 2.7.1.dev0
+- Tokenizers 0.13.2

added_tokens.json

@@ -0,0 +1,4 @@
+{
+  "</s>": 116,
+  "<s>": 115
+}

all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 75.95,
+    "eval_cer": 0.1972293657199707,
+    "eval_loss": 0.9162325859069824,
+    "eval_runtime": 45.3436,
+    "eval_samples": 481,
+    "eval_samples_per_second": 10.608,
+    "eval_steps_per_second": 0.684,
+    "eval_wer": 0.5159447476125512,
+    "train_loss": 0.044292491674423215,
+    "train_runtime": 2233.4842,
+    "train_samples": 2528,
+    "train_samples_per_second": 85.964,
+    "train_steps_per_second": 2.686
+}

config.json

@@ -0,0 +1,107 @@
+{
+  "_name_or_path": "facebook/wav2vec2-xls-r-300m",
+  "activation_dropout": 0.1,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "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": false,
+  "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": 0.0,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.0,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.1,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.3,
+  "model_type": "wav2vec2",
+  "num_adapter_layers": 3,
+  "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,
+  "output_hidden_size": 1024,
+  "pad_token_id": 114,
+  "proj_codevector_dim": 768,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.26.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 117,
+  "xvector_output_dim": 512
+}

eval.py

@@ -0,0 +1,146 @@
+'''
+python3 eval.py --model_id ./wav2vec2-xlsr-300m-pashto  --dataset google/fleurs --config ps_af --split test
+
+'''
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+import torch
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline
+
+
+def log_results(result: Dataset, args: Dict[str, str]):
+    """DO NOT CHANGE. This function computes and logs the result metrics."""
+
+    log_outputs = args.log_outputs
+    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
+
+    # load metric
+    wer = load_metric("wer")
+    cer = load_metric("cer")
+
+    # compute metrics
+    wer_result = wer.compute(
+        references=result["target"], predictions=result["prediction"])
+    cer_result = cer.compute(
+        references=result["target"], predictions=result["prediction"])
+
+    # print & log results
+    result_str = f"WER: {wer_result}\nCER: {cer_result}"
+    print(result_str)
+
+    with open(f"{dataset_id}_eval_results.txt", "w") as f:
+        f.write(result_str)
+
+    # log all results in text file. Possibly interesting for analysis
+    if log_outputs is not None:
+        pred_file = f"log_{dataset_id}_predictions.txt"
+        target_file = f"log_{dataset_id}_targets.txt"
+
+        with open(pred_file, "w") as p, open(target_file, "w") as t:
+
+            # mapping function to write output
+            def write_to_file(batch, i):
+                p.write(f"{i}" + "\n")
+                p.write(batch["prediction"] + "\n")
+                t.write(f"{i}" + "\n")
+                t.write(batch["target"] + "\n")
+
+            result.map(write_to_file, with_indices=True)
+
+
+def normalize_text(text: str) -> str:
+    """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
+
+    chars_to_ignore_regex = '[,?.!\-\;\:"“%‘”�—’…–]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    text = re.sub(chars_to_ignore_regex, "", text.lower())
+
+    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
+    # note that order is important here!
+    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]
+
+    for t in token_sequences_to_ignore:
+        text = " ".join(text.split(t))
+
+    return text
+
+
+def main(args):
+    # load dataset
+    dataset = load_dataset(args.dataset, args.config,
+                           split=args.split, use_auth_token=True)
+
+    # for testing: only process the first two examples as a test
+    # dataset = dataset.select(range(10))
+
+    # load processor
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    if args.device is None:
+        args.device = 0 if torch.cuda.is_available() else -1
+    asr = pipeline("automatic-speech-recognition",
+                   model=args.model_id, device=args.device)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        prediction = asr(
+            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
+        )
+
+        batch["prediction"] = prediction["text"]
+        batch["target"] = normalize_text(batch["transcription"])
+        return batch
+
+    # run inference on all examples
+    result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
+
+    # compute and log_results
+    # do not change function below
+    log_results(result, args)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
+    )
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        required=True,
+        help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
+    )
+    parser.add_argument(
+        "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'`  for Common Voice"
+    )
+    parser.add_argument("--split", type=str, required=True,
+                        help="Split of the dataset. *E.g.* `'test'`")
+    parser.add_argument(
+        "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
+    )
+    parser.add_argument(
+        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
+    )
+    parser.add_argument(
+        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
+    )
+    parser.add_argument(
+        "--device",
+        type=int,
+        default=None,
+        help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
+    )
+    args = parser.parse_args()
+
+    main(args)

eval_results.json

@@ -0,0 +1,10 @@
+{
+    "epoch": 75.95,
+    "eval_cer": 0.1972293657199707,
+    "eval_loss": 0.9162325859069824,
+    "eval_runtime": 45.3436,
+    "eval_samples": 481,
+    "eval_samples_per_second": 10.608,
+    "eval_steps_per_second": 0.684,
+    "eval_wer": 0.5159447476125512
+}

google_fleurs_ps_af_test_eval_results.txt

@@ -0,0 +1,2 @@
+WER: 0.5107077764277035
+CER: 0.2001802222741381

preprocessor_config.json

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

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16d5bdabf704db530a89e8f651eaa5068007117858043a754242f0b36576ecc4
+size 1262381549

run.sh

@@ -0,0 +1,33 @@
+python run_speech_recognition_ctc.py \
+      --dataset_name="google/fleurs" \
+      --dataset_config_name="ps_af" \
+      --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
+      --output_dir="./" \
+      --overwrite_output_dir \
+      --num_train_epochs="50" \
+      --per_device_train_batch_size="8" \
+      --per_device_eval_batch_size="8" \
+      --gradient_accumulation_steps="4" \
+      --learning_rate="7.5e-5" \
+      --warmup_steps="2000" \
+      --evaluation_strategy="steps" \
+      --text_column_name="transcription" \
+      --save_steps="500" \
+      --eval_steps="500" \
+      --logging_steps="10" \
+      --layerdrop="0.0" \
+      --activation_dropout="0.1" \
+      --eval_metrics wer cer \
+      --save_total_limit="1" \
+      --mask_time_prob="0.3" \
+      --mask_time_length="10" \
+      --mask_feature_prob="0.1" \
+      --fp16 \
+      --mask_feature_length="64" \
+      --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \
+      --group_by_length \
+      --push_to_hub \
+      --do_train --do_eval \
+      --gradient_checkpointing \
+      --use_auth_token 
+      --freeze_feature_extractor="True" 

run_2.sh

@@ -0,0 +1,35 @@
+python run_speech_recognition_ctc.py \
+      --dataset_name="google/fleurs" \
+      --dataset_config_name="ps_af" \
+      --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
+      --output_dir="./" \
+      --overwrite_output_dir="False" \
+      --max_steps="4500" \
+      --per_device_train_batch_size="8" \
+      --per_device_eval_batch_size="8" \
+      --gradient_accumulation_steps="4" \
+      --learning_rate="7.5e-5" \
+      --warmup_steps="2000" \
+      --evaluation_strategy="steps" \
+      --text_column_name="transcription" \
+      --save_steps="500" \
+      --eval_steps="500" \
+      --logging_steps="10" \
+      --layerdrop="0.0" \
+      --activation_dropout="0.1" \
+      --eval_metrics wer cer \
+      --greater_is_better="False" \
+      --load_best_model_at_end \
+      --save_total_limit="3" \
+      --mask_time_prob="0.3" \
+      --mask_time_length="10" \
+      --mask_feature_prob="0.1" \
+      --fp16 \
+      --mask_feature_length="64" \
+      --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \
+      --group_by_length \
+      --push_to_hub \
+      --do_train --do_eval \
+      --gradient_checkpointing \
+      --use_auth_token 
+      --freeze_feature_extractor="True" 

run_3.sh

@@ -0,0 +1,34 @@
+python run_speech_recognition_ctc.py \
+      --dataset_name="google/fleurs" \
+      --dataset_config_name="ps_af" \
+      --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
+      --output_dir="./" \
+      --overwrite_output_dir="False" \
+      --max_steps="6000" \
+      --per_device_train_batch_size="16" \
+      --per_device_eval_batch_size="16" \
+      --gradient_accumulation_steps="2" \
+      --learning_rate="7.5e-7" \
+      --warmup_steps="1000" \
+      --evaluation_strategy="steps" \
+      --text_column_name="transcription" \
+      --save_steps="500" \
+      --eval_steps="500" \
+      --logging_steps="10" \
+      --layerdrop="0.0" \
+      --activation_dropout="0.1" \
+      --eval_metrics wer cer \
+      --greater_is_better="False" \
+      --load_best_model_at_end \
+      --save_total_limit="10" \
+      --mask_time_prob="0.3" \
+      --mask_time_length="10" \
+      --mask_feature_prob="0.1" \
+      --fp16 \
+      --mask_feature_length="64" \
+      --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \
+      --group_by_length \
+      --push_to_hub \
+      --do_train --do_eval \
+      --gradient_checkpointing \
+      --use_auth_token 

run_speech_recognition_ctc.py

@@ -0,0 +1,772 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+""" Fine-tuning a 🤗 Transformers CTC model for automatic speech recognition"""
+
+import functools
+import json
+import logging
+import os
+import re
+import sys
+import warnings
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional, Union
+
+import datasets
+import numpy as np
+import torch
+from datasets import DatasetDict, load_dataset
+
+import evaluate
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoFeatureExtractor,
+    AutoModelForCTC,
+    AutoProcessor,
+    AutoTokenizer,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    Wav2Vec2Processor,
+    set_seed,
+)
+from transformers.trainer_utils import get_last_checkpoint, is_main_process
+from transformers.utils import check_min_version, send_example_telemetry
+from transformers.utils.versions import require_version
+
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.26.0.dev0")
+
+require_version("datasets>=1.18.0", "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt")
+
+
+logger = logging.getLogger(__name__)
+
+
+def list_field(default=None, metadata=None):
+    return field(default_factory=lambda: default, metadata=metadata)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    tokenizer_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={"help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"},
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
+    )
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
+    )
+    attention_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for the attention probabilities."}
+    )
+    activation_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
+    )
+    feat_proj_dropout: float = field(default=0.0, metadata={"help": "The dropout ratio for the projected features."})
+    hidden_dropout: float = field(
+        default=0.0,
+        metadata={
+            "help": "The dropout probability for all fully connected layers in the embeddings, encoder, and pooler."
+        },
+    )
+    final_dropout: float = field(
+        default=0.0,
+        metadata={"help": "The dropout probability for the final projection layer."},
+    )
+    mask_time_prob: float = field(
+        default=0.05,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the time axis to be chosen as the start of the vector"
+                "span to be masked. Approximately ``mask_time_prob * sequence_length // mask_time_length`` feature"
+                "vectors will be masked along the time axis."
+            )
+        },
+    )
+    mask_time_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the time axis."},
+    )
+    mask_feature_prob: float = field(
+        default=0.0,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the feature axis to be chosen as the start of the vectorspan"
+                " to be masked. Approximately ``mask_feature_prob * sequence_length // mask_feature_length`` feature"
+                " bins will be masked along the time axis."
+            )
+        },
+    )
+    mask_feature_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the feature axis."},
+    )
+    layerdrop: float = field(default=0.0, metadata={"help": "The LayerDrop probability."})
+    ctc_loss_reduction: Optional[str] = field(
+        default="mean", metadata={"help": "The way the ctc loss should be reduced. Should be one of 'mean' or 'sum'."}
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    dataset_name: str = field(
+        metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: str = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    train_split_name: str = field(
+        default="train+validation",
+        metadata={
+            "help": (
+                "The name of the training data set split to use (via the datasets library). Defaults to "
+                "'train+validation'"
+            )
+        },
+    )
+    eval_split_name: str = field(
+        default="test",
+        metadata={
+            "help": "The name of the evaluation data set split to use (via the datasets library). Defaults to 'test'"
+        },
+    )
+    audio_column_name: str = field(
+        default="audio",
+        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
+    )
+    text_column_name: str = field(
+        default="text",
+        metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of training examples to this "
+                "value if set."
+            )
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of validation examples to this "
+                "value if set."
+            )
+        },
+    )
+    chars_to_ignore: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to remove from the transcripts."},
+    )
+    eval_metrics: List[str] = list_field(
+        default=["wer"],
+        metadata={"help": "A list of metrics the model should be evaluated on. E.g. `'wer cer'`"},
+    )
+    max_duration_in_seconds: float = field(
+        default=20.0,
+        metadata={
+            "help": (
+                "Filter audio files that are longer than `max_duration_in_seconds` seconds to"
+                " 'max_duration_in_seconds`"
+            )
+        },
+    )
+    min_duration_in_seconds: float = field(
+        default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
+    )
+    preprocessing_only: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Whether to only do data preprocessing and skip training. This is especially useful when data"
+                " preprocessing errors out in distributed training due to timeout. In this case, one should run the"
+                " preprocessing in a non-distributed setup with `preprocessing_only=True` so that the cached datasets"
+                " can consequently be loaded in distributed training"
+            )
+        },
+    )
+    use_auth_token: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "If :obj:`True`, will use the token generated when running"
+                ":obj:`huggingface-cli login` as HTTP bearer authorization for remote files."
+            )
+        },
+    )
+    unk_token: str = field(
+        default="[UNK]",
+        metadata={"help": "The unk token for the tokenizer"},
+    )
+    pad_token: str = field(
+        default="[PAD]",
+        metadata={"help": "The padding token for the tokenizer"},
+    )
+    word_delimiter_token: str = field(
+        default="|",
+        metadata={"help": "The word delimiter token for the tokenizer"},
+    )
+    phoneme_language: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": (
+                "The target language that should be used be"
+                " passed to the tokenizer for tokenization. Note that"
+                " this is only relevant if the model classifies the"
+                " input audio to a sequence of phoneme sequences."
+            )
+        },
+    )
+
+
+@dataclass
+class DataCollatorCTCWithPadding:
+    """
+    Data collator that will dynamically pad the inputs received.
+    Args:
+        processor (:class:`~transformers.AutoProcessor`)
+            The processor used for proccessing the data.
+        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+              sequence if provided).
+            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
+              maximum acceptable input length for the model if that argument is not provided.
+            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+              different lengths).
+        max_length (:obj:`int`, `optional`):
+            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
+        max_length_labels (:obj:`int`, `optional`):
+            Maximum length of the ``labels`` returned list and optionally padding length (see above).
+        pad_to_multiple_of (:obj:`int`, `optional`):
+            If set will pad the sequence to a multiple of the provided value.
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+    """
+
+    processor: AutoProcessor
+    padding: Union[bool, str] = "longest"
+    pad_to_multiple_of: Optional[int] = None
+    pad_to_multiple_of_labels: Optional[int] = None
+
+    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
+        # split inputs and labels since they have to be of different lenghts and need
+        # different padding methods
+        input_features = [{"input_values": feature["input_values"]} for feature in features]
+        label_features = [{"input_ids": feature["labels"]} for feature in features]
+
+        batch = self.processor.pad(
+            input_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors="pt",
+        )
+
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
+
+        # replace padding with -100 to ignore loss correctly
+        labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
+
+        batch["labels"] = labels
+        if "attention_mask" in batch:
+            batch["attention_mask"] = batch["attention_mask"].to(torch.long)
+
+        return batch
+
+
+def create_vocabulary_from_data(
+    datasets: DatasetDict,
+    word_delimiter_token: Optional[str] = None,
+    unk_token: Optional[str] = None,
+    pad_token: Optional[str] = None,
+):
+    # Given training and test labels create vocabulary
+    def extract_all_chars(batch):
+        all_text = " ".join(batch["target_text"])
+        vocab = list(set(all_text))
+        return {"vocab": [vocab], "all_text": [all_text]}
+
+    vocabs = datasets.map(
+        extract_all_chars,
+        batched=True,
+        batch_size=-1,
+        keep_in_memory=True,
+        remove_columns=datasets["train"].column_names,
+    )
+
+    # take union of all unique characters in each dataset
+    vocab_set = functools.reduce(
+        lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]), vocabs.values()
+    )
+
+    vocab_dict = {v: k for k, v in enumerate(sorted(list(vocab_set)))}
+
+    # replace white space with delimiter token
+    if word_delimiter_token is not None:
+        vocab_dict[word_delimiter_token] = vocab_dict[" "]
+        del vocab_dict[" "]
+
+    # add unk and pad token
+    if unk_token is not None:
+        vocab_dict[unk_token] = len(vocab_dict)
+
+    if pad_token is not None:
+        vocab_dict[pad_token] = len(vocab_dict)
+
+    return vocab_dict
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_speech_recognition_ctc", model_args, data_args)
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # 1. First, let's load the dataset
+    raw_datasets = DatasetDict()
+
+    if training_args.do_train:
+        raw_datasets["train"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.train_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+
+        if data_args.audio_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'."
+                " Make sure to set `--audio_column_name` to the correct audio column - one of"
+                f" {', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.text_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--text_column_name {data_args.text_column_name} not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--text_column_name` to the correct text column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.max_train_samples is not None:
+            raw_datasets["train"] = raw_datasets["train"].select(range(data_args.max_train_samples))
+
+    if training_args.do_eval:
+        raw_datasets["eval"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.eval_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+
+        if data_args.max_eval_samples is not None:
+            raw_datasets["eval"] = raw_datasets["eval"].select(range(data_args.max_eval_samples))
+
+    # 2. We remove some special characters from the datasets
+    # that make training complicated and do not help in transcribing the speech
+    # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
+    # that could be easily picked up by the model
+    chars_to_ignore_regex = (
+        f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
+    )
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                try:
+                    os.remove(vocab_file)
+                except OSError:
+                    # in shared file-systems it might be the case that
+                    # two processes try to delete the vocab file at the some time
+                    pass
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                vocab_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "feat_proj_dropout": model_args.feat_proj_dropout,
+            "attention_dropout": model_args.attention_dropout,
+            "hidden_dropout": model_args.hidden_dropout,
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "activation_dropout": model_args.activation_dropout,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+    )
+
+    # freeze encoder
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
+
+    # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
+    # Thankfully, `datasets` takes care of automatically loading and resampling the audio,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
+    phoneme_language = data_args.phoneme_language
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        additional_kwargs = {}
+        if phoneme_language is not None:
+            additional_kwargs["phonemizer_lang"] = phoneme_language
+
+        batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: evaluate.load(metric) for metric in data_args.eval_metrics}
+
+    # for large datasets it is advised to run the preprocessing on a
+    # single machine first with ``args.preprocessing_only`` since there will mostly likely
+    # be a timeout when running the script in distributed mode.
+    # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
+    # cached dataset
+    if data_args.preprocessing_only:
+        logger.info(f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}")
+        return
+
+    def compute_metrics(pred):
+        pred_logits = pred.predictions
+        pred_ids = np.argmax(pred_logits, axis=-1)
+
+        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
+
+        pred_str = tokenizer.batch_decode(pred_ids)
+        # we do not want to group tokens when computing the metrics
+        label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
+
+        metrics = {k: v.compute(predictions=pred_str, references=label_str) for k, v in eval_metrics.items()}
+
+        return metrics
+
+    # Now save everything to be able to create a single processor later
+    if is_main_process(training_args.local_rank):
+        # save feature extractor, tokenizer and config
+        feature_extractor.save_pretrained(training_args.output_dir)
+        tokenizer.save_pretrained(training_args.output_dir)
+        config.save_pretrained(training_args.output_dir)
+
+    try:
+        processor = AutoProcessor.from_pretrained(training_args.output_dir)
+    except (OSError, KeyError):
+        warnings.warn(
+            "Loading a processor from a feature extractor config that does not"
+            " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
+            " attribute to your `preprocessor_config.json` file to suppress this warning: "
+            " `'processor_class': 'Wav2Vec2Processor'`",
+            FutureWarning,
+        )
+        processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
+
+    # Instantiate custom data collator
+    data_collator = DataCollatorCTCWithPadding(processor=processor)
+
+    # Initialize Trainer
+    trainer = Trainer(
+        model=model,
+        data_collator=data_collator,
+        args=training_args,
+        compute_metrics=compute_metrics,
+        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
+        tokenizer=feature_extractor,
+    )
+
+    # 8. Finally, we can start training
+
+    # Training
+    if training_args.do_train:
+
+        # use last checkpoint if exist
+        if last_checkpoint is not None:
+            checkpoint = last_checkpoint
+        elif os.path.isdir(model_args.model_name_or_path):
+            checkpoint = model_args.model_name_or_path
+        else:
+            checkpoint = None
+
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
+        trainer.save_model()
+
+        metrics = train_result.metrics
+        max_train_samples = (
+            data_args.max_train_samples
+            if data_args.max_train_samples is not None
+            else len(vectorized_datasets["train"])
+        )
+        metrics["train_samples"] = min(max_train_samples, len(vectorized_datasets["train"]))
+
+        trainer.log_metrics("train", metrics)
+        trainer.save_metrics("train", metrics)
+        trainer.save_state()
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+        metrics = trainer.evaluate()
+        max_eval_samples = (
+            data_args.max_eval_samples if data_args.max_eval_samples is not None else len(vectorized_datasets["eval"])
+        )
+        metrics["eval_samples"] = min(max_eval_samples, len(vectorized_datasets["eval"]))
+
+        trainer.log_metrics("eval", metrics)
+        trainer.save_metrics("eval", metrics)
+
+    # Write model card and (optionally) push to hub
+    config_name = data_args.dataset_config_name if data_args.dataset_config_name is not None else "na"
+    kwargs = {
+        "finetuned_from": model_args.model_name_or_path,
+        "tasks": "automatic-speech-recognition",
+        "tags": ["automatic-speech-recognition", data_args.dataset_name],
+        "dataset_args": (
+            f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split:"
+            f" {data_args.eval_split_name}"
+        ),
+        "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
+    }
+    if "common_voice" in data_args.dataset_name:
+        kwargs["language"] = config_name
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+    return results
+
+
+if __name__ == "__main__":
+    main()

runs/Dec14_13-29-56_129-213-22-31/1671025286.7583845/events.out.tfevents.1671025286.129-213-22-31.83694.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84fe7f9eb9bdee3f0ce972e8ae5cd67b151f661955efd2d0d8e279a1d976c2b6
+size 5629

runs/Dec14_13-29-56_129-213-22-31/events.out.tfevents.1671025286.129-213-22-31.83694.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:868a191ddd017bc677b5d7108d90ae2c4e031d5cf34f17cf5c6cb23be7122dac
+size 70052

runs/Dec14_13-29-56_129-213-22-31/events.out.tfevents.1671041124.129-213-22-31.83694.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56ec8495be91f4bd3dbb31e66bba57037df719175e531b261efc1d005b8ec6bf
+size 405

runs/Dec16_13-55-02_129-146-104-29/1671199202.2565184/events.out.tfevents.1671199202.129-146-104-29.128095.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:414238c9cc401aa54440b67cd9afd1482b1e3d03117800b010df72223777e4b6
+size 5633

runs/Dec16_13-55-02_129-146-104-29/events.out.tfevents.1671199202.129-146-104-29.128095.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de34b04e2c5880b8861a0e14b0100a15eb8fd48621e0e131dc82e4fe1d0301a7
+size 13738

runs/Dec16_13-55-02_129-146-104-29/events.out.tfevents.1671201437.129-146-104-29.128095.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08206f99f4f7b3fbfa384e19ec6ae571dfe46052f6556baacc5e7735d060a873
+size 405

runs/Dec16_14-39-42_129-146-104-29/1671201754.79521/events.out.tfevents.1671201754.129-146-104-29.129288.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd49e0eafc8f61728599067cf82cdb4c8e2dfc804418f88a9bac0a108f945f5a
+size 5633

runs/Dec16_14-39-42_129-146-104-29/events.out.tfevents.1671201754.129-146-104-29.129288.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:519a0f832341b452e5e021bcb75ae3c6153d9aa9caaa5a99261ad7d742e5e630
+size 13738

runs/Dec16_14-39-42_129-146-104-29/events.out.tfevents.1671204006.129-146-104-29.129288.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80d4d15418295c572aa4591c39561da14f6b9f84af7cb628829d3e04efcf370f
+size 405

runs/Dec16_15-29-40_129-146-104-29/1671204751.2903225/events.out.tfevents.1671204751.129-146-104-29.131453.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c70656d64c1d71a68951d71fc889f492d33a57405e8d161c35d2cbb6085a13e4
+size 5633

runs/Dec16_15-29-40_129-146-104-29/events.out.tfevents.1671204751.129-146-104-29.131453.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dd4b02386c90cfc08d2d69cfaf6701c41bf670404378038bb0a43b850490b37
+size 5642

runs/Dec16_15-39-46_129-146-104-29/1671205356.7546594/events.out.tfevents.1671205356.129-146-104-29.131763.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0275113dba0a656e712b957344ac77a8789fbfa165be78d96631e171b0746187
+size 5633

runs/Dec16_15-39-46_129-146-104-29/events.out.tfevents.1671205356.129-146-104-29.131763.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a08d8d7c897ed3b076f8d828040d3477a13578a5a86f64356debf482e5da28d
+size 13738

runs/Dec16_15-39-46_129-146-104-29/events.out.tfevents.1671207700.129-146-104-29.131763.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f83a25a085a2175f629e030fb53b3c3b594db36c16ed1ac8ad53615b96a60088
+size 405

runs/Dec16_20-12-50_129-146-104-29/1671221741.6851091/events.out.tfevents.1671221741.129-146-104-29.144289.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8a651984c4fb45e3bc756b5f150fcfd878800854f2d3260816244f2d70ec758
+size 5633

runs/Dec16_20-12-50_129-146-104-29/events.out.tfevents.1671221741.129-146-104-29.144289.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c94b3ebe03452e093f885924b88ff8bb7938aee4cf25e9deffade7b9711bb853
+size 13738

runs/Dec16_20-12-50_129-146-104-29/events.out.tfevents.1671223965.129-146-104-29.144289.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80e8f010a29167816de5acc6dfa4ebc9166999dc4d6436f409cabb8252b3b9dc
+size 405

runs/Dec16_20-56-58_129-146-104-29/1671224389.6246047/events.out.tfevents.1671224389.129-146-104-29.146388.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61e148fd7677c43d22f7c5ef9aad820a1b1d94af7c61b0bd4e3bbb4160ca62cd
+size 5633

runs/Dec16_20-56-58_129-146-104-29/events.out.tfevents.1671224389.129-146-104-29.146388.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15a413eb22f0a56eddd96a1300a9c14d3c643429ce13383689d03bf93592f4df
+size 6114

runs/Dec16_21-09-39_129-146-104-29/1671225152.348097/events.out.tfevents.1671225152.129-146-104-29.146624.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:312a6a0e0842bed6b7a490b5edd74a368578cf4383631f88c7e50d37d8a93214
+size 5633

runs/Dec16_21-09-39_129-146-104-29/events.out.tfevents.1671225152.129-146-104-29.146624.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f83beec22f9ef84db038caf7ebeecbf3750ce0e6139f7056f09119fed66b249
+size 13741

runs/Dec16_21-09-39_129-146-104-29/events.out.tfevents.1671227491.129-146-104-29.146624.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40319fc688859220e783165235dc6652e723535f892fdb147887467e472c1a18
+size 405

special_tokens_map.json

@@ -0,0 +1,120 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "[PAD]",
+  "unk_token": "[UNK]"
+}

tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "./",
+  "pad_token": "[PAD]",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "unk_token": "[UNK]",
+  "word_delimiter_token": "|"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 75.95,
+    "train_loss": 0.044292491674423215,
+    "train_runtime": 2233.4842,
+    "train_samples": 2528,
+    "train_samples_per_second": 85.964,
+    "train_steps_per_second": 2.686
+}

trainer_state.json

@@ -0,0 +1,3745 @@
+{
+  "best_metric": 0.9162458181381226,
+  "best_model_checkpoint": "./checkpoint-4000",
+  "epoch": 75.9493670886076,
+  "global_step": 6000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.7499999999999996e-07,
+      "loss": 23.4903,
+      "step": 10
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.499999999999999e-07,
+      "loss": 21.1248,
+      "step": 20
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.1249999999999998e-06,
+      "loss": 22.9317,
+      "step": 30
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.4625e-06,
+      "loss": 20.6205,
+      "step": 40
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.8375e-06,
+      "loss": 22.2659,
+      "step": 50
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.2124999999999996e-06,
+      "loss": 21.4276,
+      "step": 60
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 2.5875e-06,
+      "loss": 21.8665,
+      "step": 70
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 2.9624999999999996e-06,
+      "loss": 20.8487,
+      "step": 80
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 3.3374999999999994e-06,
+      "loss": 21.838,
+      "step": 90
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 3.7125e-06,
+      "loss": 19.0875,
+      "step": 100
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.087499999999999e-06,
+      "loss": 19.3293,
+      "step": 110
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 4.462499999999999e-06,
+      "loss": 16.2192,
+      "step": 120
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 4.8375e-06,
+      "loss": 15.0126,
+      "step": 130
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 5.2125e-06,
+      "loss": 13.5756,
+      "step": 140
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 5.5874999999999994e-06,
+      "loss": 12.5467,
+      "step": 150
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 5.962499999999999e-06,
+      "loss": 11.8743,
+      "step": 160
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 6.3375e-06,
+      "loss": 11.4262,
+      "step": 170
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 6.712499999999999e-06,
+      "loss": 10.3319,
+      "step": 180
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 7.0874999999999995e-06,
+      "loss": 10.0354,
+      "step": 190
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 7.4625e-06,
+      "loss": 9.8568,
+      "step": 200
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 7.837499999999999e-06,
+      "loss": 9.4804,
+      "step": 210
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 8.2125e-06,
+      "loss": 9.0891,
+      "step": 220
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 8.5875e-06,
+      "loss": 8.6768,
+      "step": 230
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 8.9625e-06,
+      "loss": 8.5948,
+      "step": 240
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 9.3375e-06,
+      "loss": 8.4701,
+      "step": 250
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 9.712499999999999e-06,
+      "loss": 8.2693,
+      "step": 260
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 1.00875e-05,
+      "loss": 8.0611,
+      "step": 270
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 1.04625e-05,
+      "loss": 8.0222,
+      "step": 280
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 1.0837499999999997e-05,
+      "loss": 7.5698,
+      "step": 290
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 1.1212499999999998e-05,
+      "loss": 7.5901,
+      "step": 300
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 1.1587499999999999e-05,
+      "loss": 7.2712,
+      "step": 310
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1.19625e-05,
+      "loss": 7.4485,
+      "step": 320
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 1.23375e-05,
+      "loss": 7.1484,
+      "step": 330
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 1.2712499999999999e-05,
+      "loss": 7.1105,
+      "step": 340
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 1.3087499999999998e-05,
+      "loss": 6.8399,
+      "step": 350
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 1.3462499999999999e-05,
+      "loss": 6.8718,
+      "step": 360
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 1.38375e-05,
+      "loss": 6.5045,
+      "step": 370
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 1.4212499999999998e-05,
+      "loss": 6.574,
+      "step": 380
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 1.4587499999999999e-05,
+      "loss": 6.1716,
+      "step": 390
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 1.49625e-05,
+      "loss": 6.3618,
+      "step": 400
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 1.5337499999999997e-05,
+      "loss": 6.012,
+      "step": 410
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 1.57125e-05,
+      "loss": 6.0979,
+      "step": 420
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 1.6087499999999998e-05,
+      "loss": 5.7887,
+      "step": 430
+    },
+    {
+      "epoch": 5.57,
+      "learning_rate": 1.6462499999999997e-05,
+      "loss": 5.892,
+      "step": 440
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 1.68375e-05,
+      "loss": 5.4672,
+      "step": 450
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 1.7212499999999998e-05,
+      "loss": 5.6353,
+      "step": 460
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 1.7587499999999997e-05,
+      "loss": 5.2754,
+      "step": 470
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 1.7962499999999996e-05,
+      "loss": 5.3343,
+      "step": 480
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 1.83375e-05,
+      "loss": 5.0586,
+      "step": 490
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 1.8712499999999997e-05,
+      "loss": 5.0767,
+      "step": 500
+    },
+    {
+      "epoch": 6.33,
+      "eval_cer": 1.0,
+      "eval_loss": 4.878269672393799,
+      "eval_runtime": 44.5452,
+      "eval_samples_per_second": 10.798,
+      "eval_steps_per_second": 1.369,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 6.46,
+      "learning_rate": 1.90875e-05,
+      "loss": 4.7637,
+      "step": 510
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 1.94625e-05,
+      "loss": 4.8495,
+      "step": 520
+    },
+    {
+      "epoch": 6.71,
+      "learning_rate": 1.9837499999999998e-05,
+      "loss": 4.5651,
+      "step": 530
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 2.02125e-05,
+      "loss": 4.7084,
+      "step": 540
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 2.05875e-05,
+      "loss": 4.3947,
+      "step": 550
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 2.09625e-05,
+      "loss": 4.4641,
+      "step": 560
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 2.1337499999999997e-05,
+      "loss": 4.2175,
+      "step": 570
+    },
+    {
+      "epoch": 7.34,
+      "learning_rate": 2.1712499999999996e-05,
+      "loss": 4.2767,
+      "step": 580
+    },
+    {
+      "epoch": 7.47,
+      "learning_rate": 2.2087499999999998e-05,
+      "loss": 4.0216,
+      "step": 590
+    },
+    {
+      "epoch": 7.59,
+      "learning_rate": 2.2462499999999997e-05,
+      "loss": 4.0459,
+      "step": 600
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 2.2837499999999996e-05,
+      "loss": 3.8726,
+      "step": 610
+    },
+    {
+      "epoch": 7.85,
+      "learning_rate": 2.32125e-05,
+      "loss": 3.8982,
+      "step": 620
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 2.3587499999999997e-05,
+      "loss": 3.7419,
+      "step": 630
+    },
+    {
+      "epoch": 8.1,
+      "learning_rate": 2.39625e-05,
+      "loss": 3.7623,
+      "step": 640
+    },
+    {
+      "epoch": 8.23,
+      "learning_rate": 2.43375e-05,
+      "loss": 3.6173,
+      "step": 650
+    },
+    {
+      "epoch": 8.35,
+      "learning_rate": 2.4712499999999998e-05,
+      "loss": 3.625,
+      "step": 660
+    },
+    {
+      "epoch": 8.48,
+      "learning_rate": 2.50875e-05,
+      "loss": 3.4981,
+      "step": 670
+    },
+    {
+      "epoch": 8.61,
+      "learning_rate": 2.54625e-05,
+      "loss": 3.5114,
+      "step": 680
+    },
+    {
+      "epoch": 8.73,
+      "learning_rate": 2.5837499999999994e-05,
+      "loss": 3.42,
+      "step": 690
+    },
+    {
+      "epoch": 8.86,
+      "learning_rate": 2.6212499999999997e-05,
+      "loss": 3.4173,
+      "step": 700
+    },
+    {
+      "epoch": 8.99,
+      "learning_rate": 2.6587499999999996e-05,
+      "loss": 3.3501,
+      "step": 710
+    },
+    {
+      "epoch": 9.11,
+      "learning_rate": 2.6962499999999998e-05,
+      "loss": 3.3289,
+      "step": 720
+    },
+    {
+      "epoch": 9.24,
+      "learning_rate": 2.7337499999999997e-05,
+      "loss": 3.2786,
+      "step": 730
+    },
+    {
+      "epoch": 9.37,
+      "learning_rate": 2.7712499999999996e-05,
+      "loss": 3.2836,
+      "step": 740
+    },
+    {
+      "epoch": 9.49,
+      "learning_rate": 2.80875e-05,
+      "loss": 3.2514,
+      "step": 750
+    },
+    {
+      "epoch": 9.62,
+      "learning_rate": 2.8462499999999997e-05,
+      "loss": 3.2778,
+      "step": 760
+    },
+    {
+      "epoch": 9.75,
+      "learning_rate": 2.88375e-05,
+      "loss": 3.2042,
+      "step": 770
+    },
+    {
+      "epoch": 9.87,
+      "learning_rate": 2.92125e-05,
+      "loss": 3.1964,
+      "step": 780
+    },
+    {
+      "epoch": 10.0,
+      "learning_rate": 2.9587499999999998e-05,
+      "loss": 3.2052,
+      "step": 790
+    },
+    {
+      "epoch": 10.13,
+      "learning_rate": 2.99625e-05,
+      "loss": 3.1989,
+      "step": 800
+    },
+    {
+      "epoch": 10.25,
+      "learning_rate": 3.03375e-05,
+      "loss": 3.1823,
+      "step": 810
+    },
+    {
+      "epoch": 10.38,
+      "learning_rate": 3.0712499999999994e-05,
+      "loss": 3.1613,
+      "step": 820
+    },
+    {
+      "epoch": 10.51,
+      "learning_rate": 3.10875e-05,
+      "loss": 3.1659,
+      "step": 830
+    },
+    {
+      "epoch": 10.63,
+      "learning_rate": 3.14625e-05,
+      "loss": 3.1798,
+      "step": 840
+    },
+    {
+      "epoch": 10.76,
+      "learning_rate": 3.1837499999999995e-05,
+      "loss": 3.1711,
+      "step": 850
+    },
+    {
+      "epoch": 10.89,
+      "learning_rate": 3.22125e-05,
+      "loss": 3.1659,
+      "step": 860
+    },
+    {
+      "epoch": 11.01,
+      "learning_rate": 3.25875e-05,
+      "loss": 3.1309,
+      "step": 870
+    },
+    {
+      "epoch": 11.14,
+      "learning_rate": 3.2962499999999995e-05,
+      "loss": 3.1706,
+      "step": 880
+    },
+    {
+      "epoch": 11.27,
+      "learning_rate": 3.33375e-05,
+      "loss": 3.1441,
+      "step": 890
+    },
+    {
+      "epoch": 11.39,
+      "learning_rate": 3.37125e-05,
+      "loss": 3.1341,
+      "step": 900
+    },
+    {
+      "epoch": 11.52,
+      "learning_rate": 3.4087499999999995e-05,
+      "loss": 3.1594,
+      "step": 910
+    },
+    {
+      "epoch": 11.65,
+      "learning_rate": 3.44625e-05,
+      "loss": 3.1262,
+      "step": 920
+    },
+    {
+      "epoch": 11.77,
+      "learning_rate": 3.48375e-05,
+      "loss": 3.1541,
+      "step": 930
+    },
+    {
+      "epoch": 11.9,
+      "learning_rate": 3.5212499999999995e-05,
+      "loss": 3.1393,
+      "step": 940
+    },
+    {
+      "epoch": 12.03,
+      "learning_rate": 3.55875e-05,
+      "loss": 3.1398,
+      "step": 950
+    },
+    {
+      "epoch": 12.15,
+      "learning_rate": 3.596249999999999e-05,
+      "loss": 3.1551,
+      "step": 960
+    },
+    {
+      "epoch": 12.28,
+      "learning_rate": 3.6337499999999996e-05,
+      "loss": 3.1391,
+      "step": 970
+    },
+    {
+      "epoch": 12.41,
+      "learning_rate": 3.67125e-05,
+      "loss": 3.1236,
+      "step": 980
+    },
+    {
+      "epoch": 12.53,
+      "learning_rate": 3.7087499999999993e-05,
+      "loss": 3.1327,
+      "step": 990
+    },
+    {
+      "epoch": 12.66,
+      "learning_rate": 3.7462499999999996e-05,
+      "loss": 3.1156,
+      "step": 1000
+    },
+    {
+      "epoch": 12.66,
+      "eval_cer": 1.0,
+      "eval_loss": 3.0990264415740967,
+      "eval_runtime": 44.9243,
+      "eval_samples_per_second": 10.707,
+      "eval_steps_per_second": 1.358,
+      "eval_wer": 1.0,
+      "step": 1000
+    },
+    {
+      "epoch": 12.78,
+      "learning_rate": 3.783749999999999e-05,
+      "loss": 3.1365,
+      "step": 1010
+    },
+    {
+      "epoch": 12.91,
+      "learning_rate": 3.8212499999999994e-05,
+      "loss": 3.1123,
+      "step": 1020
+    },
+    {
+      "epoch": 13.04,
+      "learning_rate": 3.8587499999999996e-05,
+      "loss": 3.132,
+      "step": 1030
+    },
+    {
+      "epoch": 13.16,
+      "learning_rate": 3.896249999999999e-05,
+      "loss": 3.1218,
+      "step": 1040
+    },
+    {
+      "epoch": 13.29,
+      "learning_rate": 3.9337499999999994e-05,
+      "loss": 3.1266,
+      "step": 1050
+    },
+    {
+      "epoch": 13.42,
+      "learning_rate": 3.9712499999999996e-05,
+      "loss": 3.1247,
+      "step": 1060
+    },
+    {
+      "epoch": 13.54,
+      "learning_rate": 4.008749999999999e-05,
+      "loss": 3.1203,
+      "step": 1070
+    },
+    {
+      "epoch": 13.67,
+      "learning_rate": 4.0462499999999994e-05,
+      "loss": 3.1071,
+      "step": 1080
+    },
+    {
+      "epoch": 13.8,
+      "learning_rate": 4.0837499999999997e-05,
+      "loss": 3.1095,
+      "step": 1090
+    },
+    {
+      "epoch": 13.92,
+      "learning_rate": 4.12125e-05,
+      "loss": 3.0724,
+      "step": 1100
+    },
+    {
+      "epoch": 14.05,
+      "learning_rate": 4.1587499999999994e-05,
+      "loss": 3.1,
+      "step": 1110
+    },
+    {
+      "epoch": 14.18,
+      "learning_rate": 4.19625e-05,
+      "loss": 3.0862,
+      "step": 1120
+    },
+    {
+      "epoch": 14.3,
+      "learning_rate": 4.23375e-05,
+      "loss": 3.1141,
+      "step": 1130
+    },
+    {
+      "epoch": 14.43,
+      "learning_rate": 4.2712499999999995e-05,
+      "loss": 3.0847,
+      "step": 1140
+    },
+    {
+      "epoch": 14.56,
+      "learning_rate": 4.30875e-05,
+      "loss": 3.0845,
+      "step": 1150
+    },
+    {
+      "epoch": 14.68,
+      "learning_rate": 4.34625e-05,
+      "loss": 3.0537,
+      "step": 1160
+    },
+    {
+      "epoch": 14.81,
+      "learning_rate": 4.3837499999999995e-05,
+      "loss": 3.0811,
+      "step": 1170
+    },
+    {
+      "epoch": 14.94,
+      "learning_rate": 4.42125e-05,
+      "loss": 3.031,
+      "step": 1180
+    },
+    {
+      "epoch": 15.06,
+      "learning_rate": 4.45875e-05,
+      "loss": 3.0431,
+      "step": 1190
+    },
+    {
+      "epoch": 15.19,
+      "learning_rate": 4.4962499999999995e-05,
+      "loss": 2.9891,
+      "step": 1200
+    },
+    {
+      "epoch": 15.32,
+      "learning_rate": 4.53375e-05,
+      "loss": 2.9511,
+      "step": 1210
+    },
+    {
+      "epoch": 15.44,
+      "learning_rate": 4.57125e-05,
+      "loss": 2.8874,
+      "step": 1220
+    },
+    {
+      "epoch": 15.57,
+      "learning_rate": 4.60875e-05,
+      "loss": 2.8216,
+      "step": 1230
+    },
+    {
+      "epoch": 15.7,
+      "learning_rate": 4.64625e-05,
+      "loss": 2.7211,
+      "step": 1240
+    },
+    {
+      "epoch": 15.82,
+      "learning_rate": 4.68375e-05,
+      "loss": 2.6755,
+      "step": 1250
+    },
+    {
+      "epoch": 15.95,
+      "learning_rate": 4.721249999999999e-05,
+      "loss": 2.5301,
+      "step": 1260
+    },
+    {
+      "epoch": 16.08,
+      "learning_rate": 4.758749999999999e-05,
+      "loss": 2.4484,
+      "step": 1270
+    },
+    {
+      "epoch": 16.2,
+      "learning_rate": 4.7962499999999994e-05,
+      "loss": 2.2522,
+      "step": 1280
+    },
+    {
+      "epoch": 16.33,
+      "learning_rate": 4.8337499999999996e-05,
+      "loss": 2.1895,
+      "step": 1290
+    },
+    {
+      "epoch": 16.46,
+      "learning_rate": 4.871249999999999e-05,
+      "loss": 2.0274,
+      "step": 1300
+    },
+    {
+      "epoch": 16.58,
+      "learning_rate": 4.9087499999999994e-05,
+      "loss": 1.9528,
+      "step": 1310
+    },
+    {
+      "epoch": 16.71,
+      "learning_rate": 4.9462499999999996e-05,
+      "loss": 1.8849,
+      "step": 1320
+    },
+    {
+      "epoch": 16.84,
+      "learning_rate": 4.983749999999999e-05,
+      "loss": 1.8432,
+      "step": 1330
+    },
+    {
+      "epoch": 16.96,
+      "learning_rate": 5.0212499999999994e-05,
+      "loss": 1.7181,
+      "step": 1340
+    },
+    {
+      "epoch": 17.09,
+      "learning_rate": 5.0587499999999996e-05,
+      "loss": 1.7099,
+      "step": 1350
+    },
+    {
+      "epoch": 17.22,
+      "learning_rate": 5.096249999999999e-05,
+      "loss": 1.6276,
+      "step": 1360
+    },
+    {
+      "epoch": 17.34,
+      "learning_rate": 5.1337499999999994e-05,
+      "loss": 1.6456,
+      "step": 1370
+    },
+    {
+      "epoch": 17.47,
+      "learning_rate": 5.1712499999999997e-05,
+      "loss": 1.5613,
+      "step": 1380
+    },
+    {
+      "epoch": 17.59,
+      "learning_rate": 5.20875e-05,
+      "loss": 1.5553,
+      "step": 1390
+    },
+    {
+      "epoch": 17.72,
+      "learning_rate": 5.2462499999999994e-05,
+      "loss": 1.4874,
+      "step": 1400
+    },
+    {
+      "epoch": 17.85,
+      "learning_rate": 5.28375e-05,
+      "loss": 1.5079,
+      "step": 1410
+    },
+    {
+      "epoch": 17.97,
+      "learning_rate": 5.32125e-05,
+      "loss": 1.4935,
+      "step": 1420
+    },
+    {
+      "epoch": 18.1,
+      "learning_rate": 5.3587499999999995e-05,
+      "loss": 1.4843,
+      "step": 1430
+    },
+    {
+      "epoch": 18.23,
+      "learning_rate": 5.39625e-05,
+      "loss": 1.416,
+      "step": 1440
+    },
+    {
+      "epoch": 18.35,
+      "learning_rate": 5.43375e-05,
+      "loss": 1.3815,
+      "step": 1450
+    },
+    {
+      "epoch": 18.48,
+      "learning_rate": 5.4712499999999995e-05,
+      "loss": 1.4005,
+      "step": 1460
+    },
+    {
+      "epoch": 18.61,
+      "learning_rate": 5.50875e-05,
+      "loss": 1.3824,
+      "step": 1470
+    },
+    {
+      "epoch": 18.73,
+      "learning_rate": 5.54625e-05,
+      "loss": 1.3038,
+      "step": 1480
+    },
+    {
+      "epoch": 18.86,
+      "learning_rate": 5.58375e-05,
+      "loss": 1.4168,
+      "step": 1490
+    },
+    {
+      "epoch": 18.99,
+      "learning_rate": 5.62125e-05,
+      "loss": 1.3506,
+      "step": 1500
+    },
+    {
+      "epoch": 18.99,
+      "eval_cer": 0.28892684849736266,
+      "eval_loss": 1.1056294441223145,
+      "eval_runtime": 44.5929,
+      "eval_samples_per_second": 10.786,
+      "eval_steps_per_second": 1.368,
+      "eval_wer": 0.7031036834924966,
+      "step": 1500
+    },
+    {
+      "epoch": 19.11,
+      "learning_rate": 5.658749999999999e-05,
+      "loss": 1.3407,
+      "step": 1510
+    },
+    {
+      "epoch": 19.24,
+      "learning_rate": 5.696249999999999e-05,
+      "loss": 1.2849,
+      "step": 1520
+    },
+    {
+      "epoch": 19.37,
+      "learning_rate": 5.733749999999999e-05,
+      "loss": 1.3141,
+      "step": 1530
+    },
+    {
+      "epoch": 19.49,
+      "learning_rate": 5.771249999999999e-05,
+      "loss": 1.2858,
+      "step": 1540
+    },
+    {
+      "epoch": 19.62,
+      "learning_rate": 5.8087499999999996e-05,
+      "loss": 1.2842,
+      "step": 1550
+    },
+    {
+      "epoch": 19.75,
+      "learning_rate": 5.846249999999999e-05,
+      "loss": 1.2537,
+      "step": 1560
+    },
+    {
+      "epoch": 19.87,
+      "learning_rate": 5.8837499999999994e-05,
+      "loss": 1.304,
+      "step": 1570
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 5.9212499999999996e-05,
+      "loss": 1.243,
+      "step": 1580
+    },
+    {
+      "epoch": 20.13,
+      "learning_rate": 5.958749999999999e-05,
+      "loss": 1.2636,
+      "step": 1590
+    },
+    {
+      "epoch": 20.25,
+      "learning_rate": 5.9962499999999994e-05,
+      "loss": 1.2239,
+      "step": 1600
+    },
+    {
+      "epoch": 20.38,
+      "learning_rate": 6.0337499999999996e-05,
+      "loss": 1.267,
+      "step": 1610
+    },
+    {
+      "epoch": 20.51,
+      "learning_rate": 6.071249999999999e-05,
+      "loss": 1.216,
+      "step": 1620
+    },
+    {
+      "epoch": 20.63,
+      "learning_rate": 6.10875e-05,
+      "loss": 1.253,
+      "step": 1630
+    },
+    {
+      "epoch": 20.76,
+      "learning_rate": 6.14625e-05,
+      "loss": 1.2137,
+      "step": 1640
+    },
+    {
+      "epoch": 20.89,
+      "learning_rate": 6.183749999999999e-05,
+      "loss": 1.1845,
+      "step": 1650
+    },
+    {
+      "epoch": 21.01,
+      "learning_rate": 6.22125e-05,
+      "loss": 1.1501,
+      "step": 1660
+    },
+    {
+      "epoch": 21.14,
+      "learning_rate": 6.25875e-05,
+      "loss": 1.1904,
+      "step": 1670
+    },
+    {
+      "epoch": 21.27,
+      "learning_rate": 6.296249999999999e-05,
+      "loss": 1.1456,
+      "step": 1680
+    },
+    {
+      "epoch": 21.39,
+      "learning_rate": 6.33375e-05,
+      "loss": 1.1798,
+      "step": 1690
+    },
+    {
+      "epoch": 21.52,
+      "learning_rate": 6.37125e-05,
+      "loss": 1.1122,
+      "step": 1700
+    },
+    {
+      "epoch": 21.65,
+      "learning_rate": 6.408749999999999e-05,
+      "loss": 1.1512,
+      "step": 1710
+    },
+    {
+      "epoch": 21.77,
+      "learning_rate": 6.44625e-05,
+      "loss": 1.1413,
+      "step": 1720
+    },
+    {
+      "epoch": 21.9,
+      "learning_rate": 6.48375e-05,
+      "loss": 1.1652,
+      "step": 1730
+    },
+    {
+      "epoch": 22.03,
+      "learning_rate": 6.521249999999999e-05,
+      "loss": 1.177,
+      "step": 1740
+    },
+    {
+      "epoch": 22.15,
+      "learning_rate": 6.55875e-05,
+      "loss": 1.1326,
+      "step": 1750
+    },
+    {
+      "epoch": 22.28,
+      "learning_rate": 6.596249999999998e-05,
+      "loss": 1.1082,
+      "step": 1760
+    },
+    {
+      "epoch": 22.41,
+      "learning_rate": 6.633749999999999e-05,
+      "loss": 1.1321,
+      "step": 1770
+    },
+    {
+      "epoch": 22.53,
+      "learning_rate": 6.671249999999999e-05,
+      "loss": 1.0721,
+      "step": 1780
+    },
+    {
+      "epoch": 22.66,
+      "learning_rate": 6.70875e-05,
+      "loss": 1.1199,
+      "step": 1790
+    },
+    {
+      "epoch": 22.78,
+      "learning_rate": 6.746249999999999e-05,
+      "loss": 1.1436,
+      "step": 1800
+    },
+    {
+      "epoch": 22.91,
+      "learning_rate": 6.783749999999999e-05,
+      "loss": 1.0941,
+      "step": 1810
+    },
+    {
+      "epoch": 23.04,
+      "learning_rate": 6.82125e-05,
+      "loss": 1.0631,
+      "step": 1820
+    },
+    {
+      "epoch": 23.16,
+      "learning_rate": 6.85875e-05,
+      "loss": 1.064,
+      "step": 1830
+    },
+    {
+      "epoch": 23.29,
+      "learning_rate": 6.896249999999999e-05,
+      "loss": 1.0739,
+      "step": 1840
+    },
+    {
+      "epoch": 23.42,
+      "learning_rate": 6.93375e-05,
+      "loss": 1.0354,
+      "step": 1850
+    },
+    {
+      "epoch": 23.54,
+      "learning_rate": 6.97125e-05,
+      "loss": 1.0343,
+      "step": 1860
+    },
+    {
+      "epoch": 23.67,
+      "learning_rate": 7.008749999999999e-05,
+      "loss": 1.0724,
+      "step": 1870
+    },
+    {
+      "epoch": 23.8,
+      "learning_rate": 7.04625e-05,
+      "loss": 1.0982,
+      "step": 1880
+    },
+    {
+      "epoch": 23.92,
+      "learning_rate": 7.08375e-05,
+      "loss": 1.065,
+      "step": 1890
+    },
+    {
+      "epoch": 24.05,
+      "learning_rate": 7.121249999999999e-05,
+      "loss": 1.0754,
+      "step": 1900
+    },
+    {
+      "epoch": 24.18,
+      "learning_rate": 7.15875e-05,
+      "loss": 1.0708,
+      "step": 1910
+    },
+    {
+      "epoch": 24.3,
+      "learning_rate": 7.19625e-05,
+      "loss": 1.0165,
+      "step": 1920
+    },
+    {
+      "epoch": 24.43,
+      "learning_rate": 7.233749999999999e-05,
+      "loss": 1.02,
+      "step": 1930
+    },
+    {
+      "epoch": 24.56,
+      "learning_rate": 7.27125e-05,
+      "loss": 1.0985,
+      "step": 1940
+    },
+    {
+      "epoch": 24.68,
+      "learning_rate": 7.30875e-05,
+      "loss": 0.9746,
+      "step": 1950
+    },
+    {
+      "epoch": 24.81,
+      "learning_rate": 7.346249999999999e-05,
+      "loss": 1.0644,
+      "step": 1960
+    },
+    {
+      "epoch": 24.94,
+      "learning_rate": 7.38375e-05,
+      "loss": 1.0104,
+      "step": 1970
+    },
+    {
+      "epoch": 25.06,
+      "learning_rate": 7.42125e-05,
+      "loss": 1.028,
+      "step": 1980
+    },
+    {
+      "epoch": 25.19,
+      "learning_rate": 7.45875e-05,
+      "loss": 1.0107,
+      "step": 1990
+    },
+    {
+      "epoch": 25.32,
+      "learning_rate": 7.49625e-05,
+      "loss": 0.9997,
+      "step": 2000
+    },
+    {
+      "epoch": 25.32,
+      "eval_cer": 0.2301165693690988,
+      "eval_loss": 0.919084906578064,
+      "eval_runtime": 44.703,
+      "eval_samples_per_second": 10.76,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5943894952251023,
+      "step": 2000
+    },
+    {
+      "epoch": 25.44,
+      "learning_rate": 7.465384615384615e-05,
+      "loss": 0.9571,
+      "step": 2010
+    },
+    {
+      "epoch": 25.57,
+      "learning_rate": 7.426923076923075e-05,
+      "loss": 0.9801,
+      "step": 2020
+    },
+    {
+      "epoch": 25.7,
+      "learning_rate": 7.388461538461538e-05,
+      "loss": 0.9779,
+      "step": 2030
+    },
+    {
+      "epoch": 25.82,
+      "learning_rate": 7.35e-05,
+      "loss": 1.0168,
+      "step": 2040
+    },
+    {
+      "epoch": 25.95,
+      "learning_rate": 7.31153846153846e-05,
+      "loss": 0.9302,
+      "step": 2050
+    },
+    {
+      "epoch": 26.08,
+      "learning_rate": 7.273076923076923e-05,
+      "loss": 0.989,
+      "step": 2060
+    },
+    {
+      "epoch": 26.2,
+      "learning_rate": 7.234615384615385e-05,
+      "loss": 0.9357,
+      "step": 2070
+    },
+    {
+      "epoch": 26.33,
+      "learning_rate": 7.196153846153846e-05,
+      "loss": 0.9858,
+      "step": 2080
+    },
+    {
+      "epoch": 26.46,
+      "learning_rate": 7.157692307692307e-05,
+      "loss": 0.9813,
+      "step": 2090
+    },
+    {
+      "epoch": 26.58,
+      "learning_rate": 7.119230769230769e-05,
+      "loss": 0.9554,
+      "step": 2100
+    },
+    {
+      "epoch": 26.71,
+      "learning_rate": 7.08076923076923e-05,
+      "loss": 0.8935,
+      "step": 2110
+    },
+    {
+      "epoch": 26.84,
+      "learning_rate": 7.042307692307692e-05,
+      "loss": 0.9955,
+      "step": 2120
+    },
+    {
+      "epoch": 26.96,
+      "learning_rate": 7.003846153846154e-05,
+      "loss": 0.9205,
+      "step": 2130
+    },
+    {
+      "epoch": 27.09,
+      "learning_rate": 6.965384615384615e-05,
+      "loss": 0.9527,
+      "step": 2140
+    },
+    {
+      "epoch": 27.22,
+      "learning_rate": 6.926923076923075e-05,
+      "loss": 0.8899,
+      "step": 2150
+    },
+    {
+      "epoch": 27.34,
+      "learning_rate": 6.888461538461538e-05,
+      "loss": 0.9594,
+      "step": 2160
+    },
+    {
+      "epoch": 27.47,
+      "learning_rate": 6.85e-05,
+      "loss": 0.9061,
+      "step": 2170
+    },
+    {
+      "epoch": 27.59,
+      "learning_rate": 6.81153846153846e-05,
+      "loss": 0.94,
+      "step": 2180
+    },
+    {
+      "epoch": 27.72,
+      "learning_rate": 6.773076923076923e-05,
+      "loss": 0.8611,
+      "step": 2190
+    },
+    {
+      "epoch": 27.85,
+      "learning_rate": 6.734615384615385e-05,
+      "loss": 0.9391,
+      "step": 2200
+    },
+    {
+      "epoch": 27.97,
+      "learning_rate": 6.696153846153846e-05,
+      "loss": 0.8905,
+      "step": 2210
+    },
+    {
+      "epoch": 28.1,
+      "learning_rate": 6.657692307692307e-05,
+      "loss": 0.888,
+      "step": 2220
+    },
+    {
+      "epoch": 28.23,
+      "learning_rate": 6.619230769230769e-05,
+      "loss": 0.8749,
+      "step": 2230
+    },
+    {
+      "epoch": 28.35,
+      "learning_rate": 6.580769230769231e-05,
+      "loss": 0.9331,
+      "step": 2240
+    },
+    {
+      "epoch": 28.48,
+      "learning_rate": 6.542307692307692e-05,
+      "loss": 0.8135,
+      "step": 2250
+    },
+    {
+      "epoch": 28.61,
+      "learning_rate": 6.503846153846154e-05,
+      "loss": 0.9121,
+      "step": 2260
+    },
+    {
+      "epoch": 28.73,
+      "learning_rate": 6.465384615384615e-05,
+      "loss": 0.859,
+      "step": 2270
+    },
+    {
+      "epoch": 28.86,
+      "learning_rate": 6.426923076923076e-05,
+      "loss": 0.8726,
+      "step": 2280
+    },
+    {
+      "epoch": 28.99,
+      "learning_rate": 6.388461538461538e-05,
+      "loss": 0.8497,
+      "step": 2290
+    },
+    {
+      "epoch": 29.11,
+      "learning_rate": 6.35e-05,
+      "loss": 0.8673,
+      "step": 2300
+    },
+    {
+      "epoch": 29.24,
+      "learning_rate": 6.31153846153846e-05,
+      "loss": 0.8349,
+      "step": 2310
+    },
+    {
+      "epoch": 29.37,
+      "learning_rate": 6.273076923076923e-05,
+      "loss": 0.8946,
+      "step": 2320
+    },
+    {
+      "epoch": 29.49,
+      "learning_rate": 6.234615384615384e-05,
+      "loss": 0.8805,
+      "step": 2330
+    },
+    {
+      "epoch": 29.62,
+      "learning_rate": 6.196153846153846e-05,
+      "loss": 0.8752,
+      "step": 2340
+    },
+    {
+      "epoch": 29.75,
+      "learning_rate": 6.157692307692307e-05,
+      "loss": 0.8197,
+      "step": 2350
+    },
+    {
+      "epoch": 29.87,
+      "learning_rate": 6.119230769230769e-05,
+      "loss": 0.8332,
+      "step": 2360
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 6.08076923076923e-05,
+      "loss": 0.7933,
+      "step": 2370
+    },
+    {
+      "epoch": 30.13,
+      "learning_rate": 6.0423076923076924e-05,
+      "loss": 0.8712,
+      "step": 2380
+    },
+    {
+      "epoch": 30.25,
+      "learning_rate": 6.003846153846153e-05,
+      "loss": 0.824,
+      "step": 2390
+    },
+    {
+      "epoch": 30.38,
+      "learning_rate": 5.965384615384615e-05,
+      "loss": 0.8158,
+      "step": 2400
+    },
+    {
+      "epoch": 30.51,
+      "learning_rate": 5.926923076923076e-05,
+      "loss": 0.8218,
+      "step": 2410
+    },
+    {
+      "epoch": 30.63,
+      "learning_rate": 5.888461538461538e-05,
+      "loss": 0.8403,
+      "step": 2420
+    },
+    {
+      "epoch": 30.76,
+      "learning_rate": 5.85e-05,
+      "loss": 0.7986,
+      "step": 2430
+    },
+    {
+      "epoch": 30.89,
+      "learning_rate": 5.8115384615384614e-05,
+      "loss": 0.8391,
+      "step": 2440
+    },
+    {
+      "epoch": 31.01,
+      "learning_rate": 5.773076923076922e-05,
+      "loss": 0.7736,
+      "step": 2450
+    },
+    {
+      "epoch": 31.14,
+      "learning_rate": 5.734615384615384e-05,
+      "loss": 0.8478,
+      "step": 2460
+    },
+    {
+      "epoch": 31.27,
+      "learning_rate": 5.696153846153846e-05,
+      "loss": 0.7728,
+      "step": 2470
+    },
+    {
+      "epoch": 31.39,
+      "learning_rate": 5.6576923076923073e-05,
+      "loss": 0.8231,
+      "step": 2480
+    },
+    {
+      "epoch": 31.52,
+      "learning_rate": 5.619230769230769e-05,
+      "loss": 0.7602,
+      "step": 2490
+    },
+    {
+      "epoch": 31.65,
+      "learning_rate": 5.58076923076923e-05,
+      "loss": 0.7838,
+      "step": 2500
+    },
+    {
+      "epoch": 31.65,
+      "eval_cer": 0.2152122088112177,
+      "eval_loss": 0.8952043056488037,
+      "eval_runtime": 45.1763,
+      "eval_samples_per_second": 10.647,
+      "eval_steps_per_second": 1.35,
+      "eval_wer": 0.555593451568895,
+      "step": 2500
+    },
+    {
+      "epoch": 31.77,
+      "learning_rate": 5.542307692307691e-05,
+      "loss": 0.8065,
+      "step": 2510
+    },
+    {
+      "epoch": 31.9,
+      "learning_rate": 5.503846153846153e-05,
+      "loss": 0.773,
+      "step": 2520
+    },
+    {
+      "epoch": 32.03,
+      "learning_rate": 5.465384615384615e-05,
+      "loss": 0.7854,
+      "step": 2530
+    },
+    {
+      "epoch": 32.15,
+      "learning_rate": 5.426923076923076e-05,
+      "loss": 0.7724,
+      "step": 2540
+    },
+    {
+      "epoch": 32.28,
+      "learning_rate": 5.3884615384615384e-05,
+      "loss": 0.7639,
+      "step": 2550
+    },
+    {
+      "epoch": 32.41,
+      "learning_rate": 5.35e-05,
+      "loss": 0.7993,
+      "step": 2560
+    },
+    {
+      "epoch": 32.53,
+      "learning_rate": 5.311538461538461e-05,
+      "loss": 0.7957,
+      "step": 2570
+    },
+    {
+      "epoch": 32.66,
+      "learning_rate": 5.273076923076922e-05,
+      "loss": 0.7686,
+      "step": 2580
+    },
+    {
+      "epoch": 32.78,
+      "learning_rate": 5.234615384615384e-05,
+      "loss": 0.8096,
+      "step": 2590
+    },
+    {
+      "epoch": 32.91,
+      "learning_rate": 5.196153846153846e-05,
+      "loss": 0.7357,
+      "step": 2600
+    },
+    {
+      "epoch": 33.04,
+      "learning_rate": 5.1576923076923074e-05,
+      "loss": 0.7674,
+      "step": 2610
+    },
+    {
+      "epoch": 33.16,
+      "learning_rate": 5.119230769230769e-05,
+      "loss": 0.7989,
+      "step": 2620
+    },
+    {
+      "epoch": 33.29,
+      "learning_rate": 5.08076923076923e-05,
+      "loss": 0.7474,
+      "step": 2630
+    },
+    {
+      "epoch": 33.42,
+      "learning_rate": 5.042307692307692e-05,
+      "loss": 0.7153,
+      "step": 2640
+    },
+    {
+      "epoch": 33.54,
+      "learning_rate": 5.0038461538461533e-05,
+      "loss": 0.7109,
+      "step": 2650
+    },
+    {
+      "epoch": 33.67,
+      "learning_rate": 4.965384615384615e-05,
+      "loss": 0.7841,
+      "step": 2660
+    },
+    {
+      "epoch": 33.8,
+      "learning_rate": 4.926923076923076e-05,
+      "loss": 0.762,
+      "step": 2670
+    },
+    {
+      "epoch": 33.92,
+      "learning_rate": 4.8884615384615385e-05,
+      "loss": 0.7414,
+      "step": 2680
+    },
+    {
+      "epoch": 34.05,
+      "learning_rate": 4.849999999999999e-05,
+      "loss": 0.7544,
+      "step": 2690
+    },
+    {
+      "epoch": 34.18,
+      "learning_rate": 4.811538461538461e-05,
+      "loss": 0.7338,
+      "step": 2700
+    },
+    {
+      "epoch": 34.3,
+      "learning_rate": 4.773076923076922e-05,
+      "loss": 0.7266,
+      "step": 2710
+    },
+    {
+      "epoch": 34.43,
+      "learning_rate": 4.7346153846153845e-05,
+      "loss": 0.7131,
+      "step": 2720
+    },
+    {
+      "epoch": 34.56,
+      "learning_rate": 4.696153846153846e-05,
+      "loss": 0.7291,
+      "step": 2730
+    },
+    {
+      "epoch": 34.68,
+      "learning_rate": 4.6576923076923074e-05,
+      "loss": 0.7051,
+      "step": 2740
+    },
+    {
+      "epoch": 34.81,
+      "learning_rate": 4.619230769230769e-05,
+      "loss": 0.7643,
+      "step": 2750
+    },
+    {
+      "epoch": 34.94,
+      "learning_rate": 4.5807692307692304e-05,
+      "loss": 0.727,
+      "step": 2760
+    },
+    {
+      "epoch": 35.06,
+      "learning_rate": 4.542307692307692e-05,
+      "loss": 0.7142,
+      "step": 2770
+    },
+    {
+      "epoch": 35.19,
+      "learning_rate": 4.5038461538461534e-05,
+      "loss": 0.7055,
+      "step": 2780
+    },
+    {
+      "epoch": 35.32,
+      "learning_rate": 4.465384615384615e-05,
+      "loss": 0.7339,
+      "step": 2790
+    },
+    {
+      "epoch": 35.44,
+      "learning_rate": 4.426923076923077e-05,
+      "loss": 0.6956,
+      "step": 2800
+    },
+    {
+      "epoch": 35.57,
+      "learning_rate": 4.3884615384615385e-05,
+      "loss": 0.7508,
+      "step": 2810
+    },
+    {
+      "epoch": 35.7,
+      "learning_rate": 4.3499999999999993e-05,
+      "loss": 0.7072,
+      "step": 2820
+    },
+    {
+      "epoch": 35.82,
+      "learning_rate": 4.311538461538461e-05,
+      "loss": 0.7103,
+      "step": 2830
+    },
+    {
+      "epoch": 35.95,
+      "learning_rate": 4.273076923076923e-05,
+      "loss": 0.6783,
+      "step": 2840
+    },
+    {
+      "epoch": 36.08,
+      "learning_rate": 4.2346153846153845e-05,
+      "loss": 0.7419,
+      "step": 2850
+    },
+    {
+      "epoch": 36.2,
+      "learning_rate": 4.196153846153846e-05,
+      "loss": 0.7091,
+      "step": 2860
+    },
+    {
+      "epoch": 36.33,
+      "learning_rate": 4.1576923076923075e-05,
+      "loss": 0.7073,
+      "step": 2870
+    },
+    {
+      "epoch": 36.46,
+      "learning_rate": 4.119230769230768e-05,
+      "loss": 0.6937,
+      "step": 2880
+    },
+    {
+      "epoch": 36.58,
+      "learning_rate": 4.0807692307692305e-05,
+      "loss": 0.756,
+      "step": 2890
+    },
+    {
+      "epoch": 36.71,
+      "learning_rate": 4.042307692307692e-05,
+      "loss": 0.6744,
+      "step": 2900
+    },
+    {
+      "epoch": 36.84,
+      "learning_rate": 4.0038461538461534e-05,
+      "loss": 0.7165,
+      "step": 2910
+    },
+    {
+      "epoch": 36.96,
+      "learning_rate": 3.9653846153846156e-05,
+      "loss": 0.6831,
+      "step": 2920
+    },
+    {
+      "epoch": 37.09,
+      "learning_rate": 3.926923076923077e-05,
+      "loss": 0.6894,
+      "step": 2930
+    },
+    {
+      "epoch": 37.22,
+      "learning_rate": 3.888461538461538e-05,
+      "loss": 0.6419,
+      "step": 2940
+    },
+    {
+      "epoch": 37.34,
+      "learning_rate": 3.8499999999999994e-05,
+      "loss": 0.7187,
+      "step": 2950
+    },
+    {
+      "epoch": 37.47,
+      "learning_rate": 3.811538461538461e-05,
+      "loss": 0.677,
+      "step": 2960
+    },
+    {
+      "epoch": 37.59,
+      "learning_rate": 3.773076923076923e-05,
+      "loss": 0.7263,
+      "step": 2970
+    },
+    {
+      "epoch": 37.72,
+      "learning_rate": 3.734615384615384e-05,
+      "loss": 0.6257,
+      "step": 2980
+    },
+    {
+      "epoch": 37.85,
+      "learning_rate": 3.696153846153846e-05,
+      "loss": 0.7051,
+      "step": 2990
+    },
+    {
+      "epoch": 37.97,
+      "learning_rate": 3.6576923076923075e-05,
+      "loss": 0.6665,
+      "step": 3000
+    },
+    {
+      "epoch": 37.97,
+      "eval_cer": 0.2016781484053836,
+      "eval_loss": 0.8907838463783264,
+      "eval_runtime": 45.6242,
+      "eval_samples_per_second": 10.543,
+      "eval_steps_per_second": 1.337,
+      "eval_wer": 0.5251534788540245,
+      "step": 3000
+    },
+    {
+      "epoch": 38.1,
+      "learning_rate": 3.619230769230769e-05,
+      "loss": 0.7016,
+      "step": 3010
+    },
+    {
+      "epoch": 38.23,
+      "learning_rate": 3.5807692307692305e-05,
+      "loss": 0.6585,
+      "step": 3020
+    },
+    {
+      "epoch": 38.35,
+      "learning_rate": 3.542307692307692e-05,
+      "loss": 0.6673,
+      "step": 3030
+    },
+    {
+      "epoch": 38.48,
+      "learning_rate": 3.5038461538461535e-05,
+      "loss": 0.6411,
+      "step": 3040
+    },
+    {
+      "epoch": 38.61,
+      "learning_rate": 3.465384615384615e-05,
+      "loss": 0.7038,
+      "step": 3050
+    },
+    {
+      "epoch": 38.73,
+      "learning_rate": 3.4269230769230765e-05,
+      "loss": 0.6458,
+      "step": 3060
+    },
+    {
+      "epoch": 38.86,
+      "learning_rate": 3.3884615384615386e-05,
+      "loss": 0.7231,
+      "step": 3070
+    },
+    {
+      "epoch": 38.99,
+      "learning_rate": 3.3499999999999994e-05,
+      "loss": 0.6495,
+      "step": 3080
+    },
+    {
+      "epoch": 39.11,
+      "learning_rate": 3.3115384615384616e-05,
+      "loss": 0.6788,
+      "step": 3090
+    },
+    {
+      "epoch": 39.24,
+      "learning_rate": 3.273076923076923e-05,
+      "loss": 0.6452,
+      "step": 3100
+    },
+    {
+      "epoch": 39.37,
+      "learning_rate": 3.2346153846153846e-05,
+      "loss": 0.7015,
+      "step": 3110
+    },
+    {
+      "epoch": 39.49,
+      "learning_rate": 3.196153846153846e-05,
+      "loss": 0.6518,
+      "step": 3120
+    },
+    {
+      "epoch": 39.62,
+      "learning_rate": 3.1576923076923076e-05,
+      "loss": 0.6757,
+      "step": 3130
+    },
+    {
+      "epoch": 39.75,
+      "learning_rate": 3.119230769230769e-05,
+      "loss": 0.6495,
+      "step": 3140
+    },
+    {
+      "epoch": 39.87,
+      "learning_rate": 3.0807692307692305e-05,
+      "loss": 0.6434,
+      "step": 3150
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 3.0423076923076924e-05,
+      "loss": 0.6132,
+      "step": 3160
+    },
+    {
+      "epoch": 40.13,
+      "learning_rate": 3.0038461538461535e-05,
+      "loss": 0.6959,
+      "step": 3170
+    },
+    {
+      "epoch": 40.25,
+      "learning_rate": 2.965384615384615e-05,
+      "loss": 0.6468,
+      "step": 3180
+    },
+    {
+      "epoch": 40.38,
+      "learning_rate": 2.926923076923077e-05,
+      "loss": 0.6681,
+      "step": 3190
+    },
+    {
+      "epoch": 40.51,
+      "learning_rate": 2.888461538461538e-05,
+      "loss": 0.6446,
+      "step": 3200
+    },
+    {
+      "epoch": 40.63,
+      "learning_rate": 2.8499999999999998e-05,
+      "loss": 0.6554,
+      "step": 3210
+    },
+    {
+      "epoch": 40.76,
+      "learning_rate": 2.8115384615384613e-05,
+      "loss": 0.6204,
+      "step": 3220
+    },
+    {
+      "epoch": 40.89,
+      "learning_rate": 2.7730769230769228e-05,
+      "loss": 0.677,
+      "step": 3230
+    },
+    {
+      "epoch": 41.01,
+      "learning_rate": 2.7346153846153843e-05,
+      "loss": 0.5961,
+      "step": 3240
+    },
+    {
+      "epoch": 41.14,
+      "learning_rate": 2.696153846153846e-05,
+      "loss": 0.665,
+      "step": 3250
+    },
+    {
+      "epoch": 41.27,
+      "learning_rate": 2.6576923076923073e-05,
+      "loss": 0.6753,
+      "step": 3260
+    },
+    {
+      "epoch": 41.39,
+      "learning_rate": 2.619230769230769e-05,
+      "loss": 0.6387,
+      "step": 3270
+    },
+    {
+      "epoch": 41.52,
+      "learning_rate": 2.5807692307692306e-05,
+      "loss": 0.6281,
+      "step": 3280
+    },
+    {
+      "epoch": 41.65,
+      "learning_rate": 2.542307692307692e-05,
+      "loss": 0.6287,
+      "step": 3290
+    },
+    {
+      "epoch": 41.77,
+      "learning_rate": 2.5038461538461536e-05,
+      "loss": 0.6413,
+      "step": 3300
+    },
+    {
+      "epoch": 41.9,
+      "learning_rate": 2.4653846153846154e-05,
+      "loss": 0.6061,
+      "step": 3310
+    },
+    {
+      "epoch": 42.03,
+      "learning_rate": 2.4269230769230765e-05,
+      "loss": 0.648,
+      "step": 3320
+    },
+    {
+      "epoch": 42.15,
+      "learning_rate": 2.3884615384615384e-05,
+      "loss": 0.5926,
+      "step": 3330
+    },
+    {
+      "epoch": 42.28,
+      "learning_rate": 2.35e-05,
+      "loss": 0.6366,
+      "step": 3340
+    },
+    {
+      "epoch": 42.41,
+      "learning_rate": 2.3115384615384614e-05,
+      "loss": 0.6625,
+      "step": 3350
+    },
+    {
+      "epoch": 42.53,
+      "learning_rate": 2.273076923076923e-05,
+      "loss": 0.634,
+      "step": 3360
+    },
+    {
+      "epoch": 42.66,
+      "learning_rate": 2.2346153846153847e-05,
+      "loss": 0.618,
+      "step": 3370
+    },
+    {
+      "epoch": 42.78,
+      "learning_rate": 2.1961538461538458e-05,
+      "loss": 0.5911,
+      "step": 3380
+    },
+    {
+      "epoch": 42.91,
+      "learning_rate": 2.1576923076923076e-05,
+      "loss": 0.5936,
+      "step": 3390
+    },
+    {
+      "epoch": 43.04,
+      "learning_rate": 2.119230769230769e-05,
+      "loss": 0.6267,
+      "step": 3400
+    },
+    {
+      "epoch": 43.16,
+      "learning_rate": 2.0807692307692303e-05,
+      "loss": 0.6123,
+      "step": 3410
+    },
+    {
+      "epoch": 43.29,
+      "learning_rate": 2.042307692307692e-05,
+      "loss": 0.6398,
+      "step": 3420
+    },
+    {
+      "epoch": 43.42,
+      "learning_rate": 2.003846153846154e-05,
+      "loss": 0.606,
+      "step": 3430
+    },
+    {
+      "epoch": 43.54,
+      "learning_rate": 1.965384615384615e-05,
+      "loss": 0.6253,
+      "step": 3440
+    },
+    {
+      "epoch": 43.67,
+      "learning_rate": 1.9269230769230766e-05,
+      "loss": 0.5847,
+      "step": 3450
+    },
+    {
+      "epoch": 43.8,
+      "learning_rate": 1.8884615384615384e-05,
+      "loss": 0.6248,
+      "step": 3460
+    },
+    {
+      "epoch": 43.92,
+      "learning_rate": 1.85e-05,
+      "loss": 0.5884,
+      "step": 3470
+    },
+    {
+      "epoch": 44.05,
+      "learning_rate": 1.8115384615384614e-05,
+      "loss": 0.6038,
+      "step": 3480
+    },
+    {
+      "epoch": 44.18,
+      "learning_rate": 1.773076923076923e-05,
+      "loss": 0.5888,
+      "step": 3490
+    },
+    {
+      "epoch": 44.3,
+      "learning_rate": 1.7346153846153844e-05,
+      "loss": 0.6265,
+      "step": 3500
+    },
+    {
+      "epoch": 44.3,
+      "eval_cer": 0.19540855592889456,
+      "eval_loss": 0.9062958359718323,
+      "eval_runtime": 44.6904,
+      "eval_samples_per_second": 10.763,
+      "eval_steps_per_second": 1.365,
+      "eval_wer": 0.5133015006821282,
+      "step": 3500
+    },
+    {
+      "epoch": 44.43,
+      "learning_rate": 1.8412499999999997e-05,
+      "loss": 0.6002,
+      "step": 3510
+    },
+    {
+      "epoch": 44.56,
+      "learning_rate": 1.8037499999999998e-05,
+      "loss": 0.6191,
+      "step": 3520
+    },
+    {
+      "epoch": 44.68,
+      "learning_rate": 1.76625e-05,
+      "loss": 0.5811,
+      "step": 3530
+    },
+    {
+      "epoch": 44.81,
+      "learning_rate": 1.72875e-05,
+      "loss": 0.6299,
+      "step": 3540
+    },
+    {
+      "epoch": 44.94,
+      "learning_rate": 1.6912499999999998e-05,
+      "loss": 0.5605,
+      "step": 3550
+    },
+    {
+      "epoch": 45.06,
+      "learning_rate": 1.65375e-05,
+      "loss": 0.6183,
+      "step": 3560
+    },
+    {
+      "epoch": 45.19,
+      "learning_rate": 1.61625e-05,
+      "loss": 0.5852,
+      "step": 3570
+    },
+    {
+      "epoch": 45.32,
+      "learning_rate": 1.5787499999999997e-05,
+      "loss": 0.594,
+      "step": 3580
+    },
+    {
+      "epoch": 45.44,
+      "learning_rate": 1.54125e-05,
+      "loss": 0.5965,
+      "step": 3590
+    },
+    {
+      "epoch": 45.57,
+      "learning_rate": 1.50375e-05,
+      "loss": 0.6005,
+      "step": 3600
+    },
+    {
+      "epoch": 45.7,
+      "learning_rate": 1.4662499999999999e-05,
+      "loss": 0.5884,
+      "step": 3610
+    },
+    {
+      "epoch": 45.82,
+      "learning_rate": 1.4287499999999998e-05,
+      "loss": 0.5884,
+      "step": 3620
+    },
+    {
+      "epoch": 45.95,
+      "learning_rate": 1.39125e-05,
+      "loss": 0.5628,
+      "step": 3630
+    },
+    {
+      "epoch": 46.08,
+      "learning_rate": 1.3537499999999999e-05,
+      "loss": 0.6339,
+      "step": 3640
+    },
+    {
+      "epoch": 46.2,
+      "learning_rate": 1.3162499999999998e-05,
+      "loss": 0.5578,
+      "step": 3650
+    },
+    {
+      "epoch": 46.33,
+      "learning_rate": 1.2787499999999999e-05,
+      "loss": 0.6239,
+      "step": 3660
+    },
+    {
+      "epoch": 46.46,
+      "learning_rate": 1.24125e-05,
+      "loss": 0.5872,
+      "step": 3670
+    },
+    {
+      "epoch": 46.58,
+      "learning_rate": 1.20375e-05,
+      "loss": 0.5697,
+      "step": 3680
+    },
+    {
+      "epoch": 46.71,
+      "learning_rate": 1.1662499999999999e-05,
+      "loss": 0.5475,
+      "step": 3690
+    },
+    {
+      "epoch": 46.84,
+      "learning_rate": 1.1287499999999998e-05,
+      "loss": 0.5979,
+      "step": 3700
+    },
+    {
+      "epoch": 46.96,
+      "learning_rate": 1.0912499999999998e-05,
+      "loss": 0.5742,
+      "step": 3710
+    },
+    {
+      "epoch": 47.09,
+      "learning_rate": 1.05375e-05,
+      "loss": 0.6054,
+      "step": 3720
+    },
+    {
+      "epoch": 47.22,
+      "learning_rate": 1.01625e-05,
+      "loss": 0.5777,
+      "step": 3730
+    },
+    {
+      "epoch": 47.34,
+      "learning_rate": 9.787499999999999e-06,
+      "loss": 0.5734,
+      "step": 3740
+    },
+    {
+      "epoch": 47.47,
+      "learning_rate": 9.412499999999998e-06,
+      "loss": 0.5322,
+      "step": 3750
+    },
+    {
+      "epoch": 47.59,
+      "learning_rate": 9.0375e-06,
+      "loss": 0.6287,
+      "step": 3760
+    },
+    {
+      "epoch": 47.72,
+      "learning_rate": 8.6625e-06,
+      "loss": 0.547,
+      "step": 3770
+    },
+    {
+      "epoch": 47.85,
+      "learning_rate": 8.2875e-06,
+      "loss": 0.6414,
+      "step": 3780
+    },
+    {
+      "epoch": 47.97,
+      "learning_rate": 7.9125e-06,
+      "loss": 0.5661,
+      "step": 3790
+    },
+    {
+      "epoch": 48.1,
+      "learning_rate": 7.537499999999999e-06,
+      "loss": 0.5893,
+      "step": 3800
+    },
+    {
+      "epoch": 48.23,
+      "learning_rate": 7.1625e-06,
+      "loss": 0.556,
+      "step": 3810
+    },
+    {
+      "epoch": 48.35,
+      "learning_rate": 6.787499999999999e-06,
+      "loss": 0.6265,
+      "step": 3820
+    },
+    {
+      "epoch": 48.48,
+      "learning_rate": 6.4125e-06,
+      "loss": 0.5644,
+      "step": 3830
+    },
+    {
+      "epoch": 48.61,
+      "learning_rate": 6.037499999999999e-06,
+      "loss": 0.6202,
+      "step": 3840
+    },
+    {
+      "epoch": 48.73,
+      "learning_rate": 5.6624999999999996e-06,
+      "loss": 0.5581,
+      "step": 3850
+    },
+    {
+      "epoch": 48.86,
+      "learning_rate": 5.287499999999999e-06,
+      "loss": 0.572,
+      "step": 3860
+    },
+    {
+      "epoch": 48.99,
+      "learning_rate": 4.9125e-06,
+      "loss": 0.5559,
+      "step": 3870
+    },
+    {
+      "epoch": 49.11,
+      "learning_rate": 4.537499999999999e-06,
+      "loss": 0.6013,
+      "step": 3880
+    },
+    {
+      "epoch": 49.24,
+      "learning_rate": 4.1624999999999995e-06,
+      "loss": 0.5498,
+      "step": 3890
+    },
+    {
+      "epoch": 49.37,
+      "learning_rate": 3.7874999999999997e-06,
+      "loss": 0.5883,
+      "step": 3900
+    },
+    {
+      "epoch": 49.49,
+      "learning_rate": 3.4124999999999995e-06,
+      "loss": 0.5777,
+      "step": 3910
+    },
+    {
+      "epoch": 49.62,
+      "learning_rate": 3.0374999999999997e-06,
+      "loss": 0.5768,
+      "step": 3920
+    },
+    {
+      "epoch": 49.75,
+      "learning_rate": 2.6624999999999995e-06,
+      "loss": 0.5603,
+      "step": 3930
+    },
+    {
+      "epoch": 49.87,
+      "learning_rate": 2.2874999999999997e-06,
+      "loss": 0.5814,
+      "step": 3940
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 1.9125e-06,
+      "loss": 0.5562,
+      "step": 3950
+    },
+    {
+      "epoch": 50.13,
+      "learning_rate": 1.5374999999999999e-06,
+      "loss": 0.5858,
+      "step": 3960
+    },
+    {
+      "epoch": 50.25,
+      "learning_rate": 1.1624999999999999e-06,
+      "loss": 0.5279,
+      "step": 3970
+    },
+    {
+      "epoch": 50.38,
+      "learning_rate": 7.875e-07,
+      "loss": 0.5734,
+      "step": 3980
+    },
+    {
+      "epoch": 50.51,
+      "learning_rate": 4.124999999999999e-07,
+      "loss": 0.5895,
+      "step": 3990
+    },
+    {
+      "epoch": 50.63,
+      "learning_rate": 3.75e-08,
+      "loss": 0.5935,
+      "step": 4000
+    },
+    {
+      "epoch": 50.63,
+      "eval_cer": 0.1969102547256584,
+      "eval_loss": 0.9162458181381226,
+      "eval_runtime": 44.8405,
+      "eval_samples_per_second": 10.727,
+      "eval_steps_per_second": 1.36,
+      "eval_wer": 0.5156036834924966,
+      "step": 4000
+    },
+    {
+      "epoch": 50.76,
+      "learning_rate": 1.4729999999999998e-05,
+      "loss": 0.543,
+      "step": 4010
+    },
+    {
+      "epoch": 50.89,
+      "learning_rate": 1.4429999999999997e-05,
+      "loss": 0.6044,
+      "step": 4020
+    },
+    {
+      "epoch": 51.01,
+      "learning_rate": 1.413e-05,
+      "loss": 0.5749,
+      "step": 4030
+    },
+    {
+      "epoch": 51.14,
+      "learning_rate": 1.383e-05,
+      "loss": 0.6171,
+      "step": 4040
+    },
+    {
+      "epoch": 51.27,
+      "learning_rate": 1.353e-05,
+      "loss": 0.5767,
+      "step": 4050
+    },
+    {
+      "epoch": 51.39,
+      "learning_rate": 1.3229999999999999e-05,
+      "loss": 0.5749,
+      "step": 4060
+    },
+    {
+      "epoch": 51.52,
+      "learning_rate": 1.2929999999999998e-05,
+      "loss": 0.565,
+      "step": 4070
+    },
+    {
+      "epoch": 51.65,
+      "learning_rate": 1.2629999999999998e-05,
+      "loss": 0.5907,
+      "step": 4080
+    },
+    {
+      "epoch": 51.77,
+      "learning_rate": 1.2329999999999999e-05,
+      "loss": 0.575,
+      "step": 4090
+    },
+    {
+      "epoch": 51.9,
+      "learning_rate": 1.2029999999999998e-05,
+      "loss": 0.5692,
+      "step": 4100
+    },
+    {
+      "epoch": 52.03,
+      "learning_rate": 1.173e-05,
+      "loss": 0.5219,
+      "step": 4110
+    },
+    {
+      "epoch": 52.15,
+      "learning_rate": 1.1429999999999999e-05,
+      "loss": 0.5535,
+      "step": 4120
+    },
+    {
+      "epoch": 52.28,
+      "learning_rate": 1.113e-05,
+      "loss": 0.5519,
+      "step": 4130
+    },
+    {
+      "epoch": 52.41,
+      "learning_rate": 1.083e-05,
+      "loss": 0.5826,
+      "step": 4140
+    },
+    {
+      "epoch": 52.53,
+      "learning_rate": 1.0529999999999999e-05,
+      "loss": 0.5472,
+      "step": 4150
+    },
+    {
+      "epoch": 52.66,
+      "learning_rate": 1.0229999999999999e-05,
+      "loss": 0.5603,
+      "step": 4160
+    },
+    {
+      "epoch": 52.78,
+      "learning_rate": 9.929999999999998e-06,
+      "loss": 0.589,
+      "step": 4170
+    },
+    {
+      "epoch": 52.91,
+      "learning_rate": 9.629999999999998e-06,
+      "loss": 0.6005,
+      "step": 4180
+    },
+    {
+      "epoch": 53.04,
+      "learning_rate": 9.329999999999999e-06,
+      "loss": 0.5844,
+      "step": 4190
+    },
+    {
+      "epoch": 53.16,
+      "learning_rate": 9.029999999999998e-06,
+      "loss": 0.5779,
+      "step": 4200
+    },
+    {
+      "epoch": 53.29,
+      "learning_rate": 8.73e-06,
+      "loss": 0.5611,
+      "step": 4210
+    },
+    {
+      "epoch": 53.42,
+      "learning_rate": 8.429999999999999e-06,
+      "loss": 0.5859,
+      "step": 4220
+    },
+    {
+      "epoch": 53.54,
+      "learning_rate": 8.129999999999998e-06,
+      "loss": 0.5906,
+      "step": 4230
+    },
+    {
+      "epoch": 53.67,
+      "learning_rate": 7.83e-06,
+      "loss": 0.5522,
+      "step": 4240
+    },
+    {
+      "epoch": 53.8,
+      "learning_rate": 7.56e-06,
+      "loss": 0.5762,
+      "step": 4250
+    },
+    {
+      "epoch": 53.92,
+      "learning_rate": 7.259999999999999e-06,
+      "loss": 0.5498,
+      "step": 4260
+    },
+    {
+      "epoch": 54.05,
+      "learning_rate": 6.959999999999999e-06,
+      "loss": 0.5752,
+      "step": 4270
+    },
+    {
+      "epoch": 54.18,
+      "learning_rate": 6.66e-06,
+      "loss": 0.5428,
+      "step": 4280
+    },
+    {
+      "epoch": 54.3,
+      "learning_rate": 6.359999999999999e-06,
+      "loss": 0.5515,
+      "step": 4290
+    },
+    {
+      "epoch": 54.43,
+      "learning_rate": 6.06e-06,
+      "loss": 0.5662,
+      "step": 4300
+    },
+    {
+      "epoch": 54.56,
+      "learning_rate": 5.759999999999999e-06,
+      "loss": 0.5916,
+      "step": 4310
+    },
+    {
+      "epoch": 54.68,
+      "learning_rate": 5.459999999999999e-06,
+      "loss": 0.537,
+      "step": 4320
+    },
+    {
+      "epoch": 54.81,
+      "learning_rate": 5.16e-06,
+      "loss": 0.5744,
+      "step": 4330
+    },
+    {
+      "epoch": 54.94,
+      "learning_rate": 4.859999999999999e-06,
+      "loss": 0.5606,
+      "step": 4340
+    },
+    {
+      "epoch": 55.06,
+      "learning_rate": 4.5599999999999995e-06,
+      "loss": 0.5855,
+      "step": 4350
+    },
+    {
+      "epoch": 55.19,
+      "learning_rate": 4.26e-06,
+      "loss": 0.5486,
+      "step": 4360
+    },
+    {
+      "epoch": 55.32,
+      "learning_rate": 3.959999999999999e-06,
+      "loss": 0.5644,
+      "step": 4370
+    },
+    {
+      "epoch": 55.44,
+      "learning_rate": 3.66e-06,
+      "loss": 0.5525,
+      "step": 4380
+    },
+    {
+      "epoch": 55.57,
+      "learning_rate": 3.3599999999999996e-06,
+      "loss": 0.6088,
+      "step": 4390
+    },
+    {
+      "epoch": 55.7,
+      "learning_rate": 3.06e-06,
+      "loss": 0.5344,
+      "step": 4400
+    },
+    {
+      "epoch": 55.82,
+      "learning_rate": 2.76e-06,
+      "loss": 0.5379,
+      "step": 4410
+    },
+    {
+      "epoch": 55.95,
+      "learning_rate": 2.46e-06,
+      "loss": 0.5204,
+      "step": 4420
+    },
+    {
+      "epoch": 56.08,
+      "learning_rate": 2.1599999999999996e-06,
+      "loss": 0.5754,
+      "step": 4430
+    },
+    {
+      "epoch": 56.2,
+      "learning_rate": 1.8599999999999998e-06,
+      "loss": 0.5507,
+      "step": 4440
+    },
+    {
+      "epoch": 56.33,
+      "learning_rate": 1.5599999999999999e-06,
+      "loss": 0.5592,
+      "step": 4450
+    },
+    {
+      "epoch": 56.46,
+      "learning_rate": 1.2599999999999998e-06,
+      "loss": 0.5396,
+      "step": 4460
+    },
+    {
+      "epoch": 56.58,
+      "learning_rate": 9.6e-07,
+      "loss": 0.579,
+      "step": 4470
+    },
+    {
+      "epoch": 56.71,
+      "learning_rate": 6.6e-07,
+      "loss": 0.545,
+      "step": 4480
+    },
+    {
+      "epoch": 56.84,
+      "learning_rate": 3.5999999999999994e-07,
+      "loss": 0.5919,
+      "step": 4490
+    },
+    {
+      "epoch": 56.96,
+      "learning_rate": 6e-08,
+      "loss": 0.5174,
+      "step": 4500
+    },
+    {
+      "epoch": 56.96,
+      "eval_cer": 0.19719182325005163,
+      "eval_loss": 0.9287102818489075,
+      "eval_runtime": 44.3461,
+      "eval_samples_per_second": 10.847,
+      "eval_steps_per_second": 1.376,
+      "eval_wer": 0.5139836289222374,
+      "step": 4500
+    },
+    {
+      "epoch": 57.09,
+      "learning_rate": 1.2299999999999999e-05,
+      "loss": 0.5852,
+      "step": 4510
+    },
+    {
+      "epoch": 57.22,
+      "learning_rate": 1.205e-05,
+      "loss": 0.5752,
+      "step": 4520
+    },
+    {
+      "epoch": 57.34,
+      "learning_rate": 1.1799999999999999e-05,
+      "loss": 0.5433,
+      "step": 4530
+    },
+    {
+      "epoch": 57.47,
+      "learning_rate": 1.155e-05,
+      "loss": 0.5648,
+      "step": 4540
+    },
+    {
+      "epoch": 57.59,
+      "learning_rate": 1.1299999999999999e-05,
+      "loss": 0.5704,
+      "step": 4550
+    },
+    {
+      "epoch": 57.72,
+      "learning_rate": 1.105e-05,
+      "loss": 0.5216,
+      "step": 4560
+    },
+    {
+      "epoch": 57.85,
+      "learning_rate": 1.0799999999999998e-05,
+      "loss": 0.5998,
+      "step": 4570
+    },
+    {
+      "epoch": 57.97,
+      "learning_rate": 1.0549999999999999e-05,
+      "loss": 0.5439,
+      "step": 4580
+    },
+    {
+      "epoch": 58.1,
+      "learning_rate": 1.03e-05,
+      "loss": 0.5679,
+      "step": 4590
+    },
+    {
+      "epoch": 58.23,
+      "learning_rate": 1.005e-05,
+      "loss": 0.5621,
+      "step": 4600
+    },
+    {
+      "epoch": 58.35,
+      "learning_rate": 9.799999999999998e-06,
+      "loss": 0.5557,
+      "step": 4610
+    },
+    {
+      "epoch": 58.48,
+      "learning_rate": 9.549999999999998e-06,
+      "loss": 0.5525,
+      "step": 4620
+    },
+    {
+      "epoch": 58.61,
+      "learning_rate": 9.299999999999999e-06,
+      "loss": 0.6033,
+      "step": 4630
+    },
+    {
+      "epoch": 58.73,
+      "learning_rate": 9.05e-06,
+      "loss": 0.5059,
+      "step": 4640
+    },
+    {
+      "epoch": 58.86,
+      "learning_rate": 8.799999999999999e-06,
+      "loss": 0.5362,
+      "step": 4650
+    },
+    {
+      "epoch": 58.99,
+      "learning_rate": 8.55e-06,
+      "loss": 0.535,
+      "step": 4660
+    },
+    {
+      "epoch": 59.11,
+      "learning_rate": 8.299999999999998e-06,
+      "loss": 0.586,
+      "step": 4670
+    },
+    {
+      "epoch": 59.24,
+      "learning_rate": 8.05e-06,
+      "loss": 0.55,
+      "step": 4680
+    },
+    {
+      "epoch": 59.37,
+      "learning_rate": 7.799999999999998e-06,
+      "loss": 0.582,
+      "step": 4690
+    },
+    {
+      "epoch": 59.49,
+      "learning_rate": 7.55e-06,
+      "loss": 0.5065,
+      "step": 4700
+    },
+    {
+      "epoch": 59.62,
+      "learning_rate": 7.299999999999999e-06,
+      "loss": 0.5715,
+      "step": 4710
+    },
+    {
+      "epoch": 59.75,
+      "learning_rate": 7.049999999999999e-06,
+      "loss": 0.5359,
+      "step": 4720
+    },
+    {
+      "epoch": 59.87,
+      "learning_rate": 6.8e-06,
+      "loss": 0.5535,
+      "step": 4730
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 6.549999999999999e-06,
+      "loss": 0.5256,
+      "step": 4740
+    },
+    {
+      "epoch": 60.13,
+      "learning_rate": 6.3e-06,
+      "loss": 0.5869,
+      "step": 4750
+    },
+    {
+      "epoch": 60.25,
+      "learning_rate": 6.05e-06,
+      "loss": 0.4978,
+      "step": 4760
+    },
+    {
+      "epoch": 60.38,
+      "learning_rate": 5.7999999999999995e-06,
+      "loss": 0.5402,
+      "step": 4770
+    },
+    {
+      "epoch": 60.51,
+      "learning_rate": 5.549999999999999e-06,
+      "loss": 0.5607,
+      "step": 4780
+    },
+    {
+      "epoch": 60.63,
+      "learning_rate": 5.3e-06,
+      "loss": 0.5583,
+      "step": 4790
+    },
+    {
+      "epoch": 60.76,
+      "learning_rate": 5.049999999999999e-06,
+      "loss": 0.5382,
+      "step": 4800
+    },
+    {
+      "epoch": 60.89,
+      "learning_rate": 4.8e-06,
+      "loss": 0.5498,
+      "step": 4810
+    },
+    {
+      "epoch": 61.01,
+      "learning_rate": 4.55e-06,
+      "loss": 0.5443,
+      "step": 4820
+    },
+    {
+      "epoch": 61.14,
+      "learning_rate": 4.2999999999999995e-06,
+      "loss": 0.5579,
+      "step": 4830
+    },
+    {
+      "epoch": 61.27,
+      "learning_rate": 4.049999999999999e-06,
+      "loss": 0.517,
+      "step": 4840
+    },
+    {
+      "epoch": 61.39,
+      "learning_rate": 3.7999999999999996e-06,
+      "loss": 0.566,
+      "step": 4850
+    },
+    {
+      "epoch": 61.52,
+      "learning_rate": 3.5499999999999995e-06,
+      "loss": 0.572,
+      "step": 4860
+    },
+    {
+      "epoch": 61.65,
+      "learning_rate": 3.2999999999999993e-06,
+      "loss": 0.5425,
+      "step": 4870
+    },
+    {
+      "epoch": 61.77,
+      "learning_rate": 3.0499999999999996e-06,
+      "loss": 0.5617,
+      "step": 4880
+    },
+    {
+      "epoch": 61.9,
+      "learning_rate": 2.8e-06,
+      "loss": 0.5352,
+      "step": 4890
+    },
+    {
+      "epoch": 62.03,
+      "learning_rate": 2.55e-06,
+      "loss": 0.5328,
+      "step": 4900
+    },
+    {
+      "epoch": 62.15,
+      "learning_rate": 2.2999999999999996e-06,
+      "loss": 0.567,
+      "step": 4910
+    },
+    {
+      "epoch": 62.28,
+      "learning_rate": 2.05e-06,
+      "loss": 0.554,
+      "step": 4920
+    },
+    {
+      "epoch": 62.41,
+      "learning_rate": 1.8e-06,
+      "loss": 0.5846,
+      "step": 4930
+    },
+    {
+      "epoch": 62.53,
+      "learning_rate": 1.5499999999999998e-06,
+      "loss": 0.5451,
+      "step": 4940
+    },
+    {
+      "epoch": 62.66,
+      "learning_rate": 1.2999999999999998e-06,
+      "loss": 0.5251,
+      "step": 4950
+    },
+    {
+      "epoch": 62.78,
+      "learning_rate": 1.05e-06,
+      "loss": 0.5373,
+      "step": 4960
+    },
+    {
+      "epoch": 62.91,
+      "learning_rate": 7.999999999999999e-07,
+      "loss": 0.5245,
+      "step": 4970
+    },
+    {
+      "epoch": 63.04,
+      "learning_rate": 5.499999999999999e-07,
+      "loss": 0.5535,
+      "step": 4980
+    },
+    {
+      "epoch": 63.16,
+      "learning_rate": 3e-07,
+      "loss": 0.545,
+      "step": 4990
+    },
+    {
+      "epoch": 63.29,
+      "learning_rate": 4.999999999999999e-08,
+      "loss": 0.5462,
+      "step": 5000
+    },
+    {
+      "epoch": 63.29,
+      "eval_cer": 0.19739830683460666,
+      "eval_loss": 0.9370450973510742,
+      "eval_runtime": 45.0983,
+      "eval_samples_per_second": 10.666,
+      "eval_steps_per_second": 1.353,
+      "eval_wer": 0.5138130968622101,
+      "step": 5000
+    },
+    {
+      "epoch": 63.42,
+      "learning_rate": 8.2e-06,
+      "loss": 0.5146,
+      "step": 5010
+    },
+    {
+      "epoch": 63.54,
+      "learning_rate": 8.033333333333333e-06,
+      "loss": 0.5334,
+      "step": 5020
+    },
+    {
+      "epoch": 63.67,
+      "learning_rate": 7.866666666666667e-06,
+      "loss": 0.5824,
+      "step": 5030
+    },
+    {
+      "epoch": 63.8,
+      "learning_rate": 7.699999999999999e-06,
+      "loss": 0.5354,
+      "step": 5040
+    },
+    {
+      "epoch": 63.92,
+      "learning_rate": 7.533333333333333e-06,
+      "loss": 0.5225,
+      "step": 5050
+    },
+    {
+      "epoch": 64.05,
+      "learning_rate": 7.366666666666666e-06,
+      "loss": 0.5296,
+      "step": 5060
+    },
+    {
+      "epoch": 64.18,
+      "learning_rate": 7.2e-06,
+      "loss": 0.525,
+      "step": 5070
+    },
+    {
+      "epoch": 64.3,
+      "learning_rate": 7.033333333333333e-06,
+      "loss": 0.5549,
+      "step": 5080
+    },
+    {
+      "epoch": 64.43,
+      "learning_rate": 6.8666666666666664e-06,
+      "loss": 0.5579,
+      "step": 5090
+    },
+    {
+      "epoch": 64.56,
+      "learning_rate": 6.699999999999999e-06,
+      "loss": 0.5527,
+      "step": 5100
+    },
+    {
+      "epoch": 64.68,
+      "learning_rate": 6.533333333333333e-06,
+      "loss": 0.5191,
+      "step": 5110
+    },
+    {
+      "epoch": 64.81,
+      "learning_rate": 6.366666666666666e-06,
+      "loss": 0.5591,
+      "step": 5120
+    },
+    {
+      "epoch": 64.94,
+      "learning_rate": 6.199999999999999e-06,
+      "loss": 0.5371,
+      "step": 5130
+    },
+    {
+      "epoch": 65.06,
+      "learning_rate": 6.033333333333333e-06,
+      "loss": 0.5527,
+      "step": 5140
+    },
+    {
+      "epoch": 65.19,
+      "learning_rate": 5.866666666666666e-06,
+      "loss": 0.5318,
+      "step": 5150
+    },
+    {
+      "epoch": 65.32,
+      "learning_rate": 5.7e-06,
+      "loss": 0.5684,
+      "step": 5160
+    },
+    {
+      "epoch": 65.44,
+      "learning_rate": 5.533333333333333e-06,
+      "loss": 0.528,
+      "step": 5170
+    },
+    {
+      "epoch": 65.57,
+      "learning_rate": 5.366666666666666e-06,
+      "loss": 0.5366,
+      "step": 5180
+    },
+    {
+      "epoch": 65.7,
+      "learning_rate": 5.199999999999999e-06,
+      "loss": 0.5482,
+      "step": 5190
+    },
+    {
+      "epoch": 65.82,
+      "learning_rate": 5.033333333333332e-06,
+      "loss": 0.5402,
+      "step": 5200
+    },
+    {
+      "epoch": 65.95,
+      "learning_rate": 4.866666666666666e-06,
+      "loss": 0.5568,
+      "step": 5210
+    },
+    {
+      "epoch": 66.08,
+      "learning_rate": 4.699999999999999e-06,
+      "loss": 0.5466,
+      "step": 5220
+    },
+    {
+      "epoch": 66.2,
+      "learning_rate": 4.533333333333333e-06,
+      "loss": 0.5353,
+      "step": 5230
+    },
+    {
+      "epoch": 66.33,
+      "learning_rate": 4.366666666666667e-06,
+      "loss": 0.5629,
+      "step": 5240
+    },
+    {
+      "epoch": 66.46,
+      "learning_rate": 4.2e-06,
+      "loss": 0.5227,
+      "step": 5250
+    },
+    {
+      "epoch": 66.58,
+      "learning_rate": 4.033333333333333e-06,
+      "loss": 0.5126,
+      "step": 5260
+    },
+    {
+      "epoch": 66.71,
+      "learning_rate": 3.866666666666666e-06,
+      "loss": 0.5173,
+      "step": 5270
+    },
+    {
+      "epoch": 66.84,
+      "learning_rate": 3.6999999999999997e-06,
+      "loss": 0.5773,
+      "step": 5280
+    },
+    {
+      "epoch": 66.96,
+      "learning_rate": 3.533333333333333e-06,
+      "loss": 0.5131,
+      "step": 5290
+    },
+    {
+      "epoch": 67.09,
+      "learning_rate": 3.3666666666666665e-06,
+      "loss": 0.5592,
+      "step": 5300
+    },
+    {
+      "epoch": 67.22,
+      "learning_rate": 3.1999999999999994e-06,
+      "loss": 0.5164,
+      "step": 5310
+    },
+    {
+      "epoch": 67.34,
+      "learning_rate": 3.033333333333333e-06,
+      "loss": 0.5166,
+      "step": 5320
+    },
+    {
+      "epoch": 67.47,
+      "learning_rate": 2.866666666666666e-06,
+      "loss": 0.5079,
+      "step": 5330
+    },
+    {
+      "epoch": 67.59,
+      "learning_rate": 2.6999999999999996e-06,
+      "loss": 0.547,
+      "step": 5340
+    },
+    {
+      "epoch": 67.72,
+      "learning_rate": 2.533333333333333e-06,
+      "loss": 0.5188,
+      "step": 5350
+    },
+    {
+      "epoch": 67.85,
+      "learning_rate": 2.3666666666666667e-06,
+      "loss": 0.5779,
+      "step": 5360
+    },
+    {
+      "epoch": 67.97,
+      "learning_rate": 2.1999999999999997e-06,
+      "loss": 0.5424,
+      "step": 5370
+    },
+    {
+      "epoch": 68.1,
+      "learning_rate": 2.033333333333333e-06,
+      "loss": 0.5307,
+      "step": 5380
+    },
+    {
+      "epoch": 68.23,
+      "learning_rate": 1.8666666666666664e-06,
+      "loss": 0.5353,
+      "step": 5390
+    },
+    {
+      "epoch": 68.35,
+      "learning_rate": 1.7e-06,
+      "loss": 0.5521,
+      "step": 5400
+    },
+    {
+      "epoch": 68.48,
+      "learning_rate": 1.5333333333333332e-06,
+      "loss": 0.5024,
+      "step": 5410
+    },
+    {
+      "epoch": 68.61,
+      "learning_rate": 1.3666666666666666e-06,
+      "loss": 0.5765,
+      "step": 5420
+    },
+    {
+      "epoch": 68.73,
+      "learning_rate": 1.2e-06,
+      "loss": 0.497,
+      "step": 5430
+    },
+    {
+      "epoch": 68.86,
+      "learning_rate": 1.0333333333333333e-06,
+      "loss": 0.5822,
+      "step": 5440
+    },
+    {
+      "epoch": 68.99,
+      "learning_rate": 8.666666666666666e-07,
+      "loss": 0.5189,
+      "step": 5450
+    },
+    {
+      "epoch": 69.11,
+      "learning_rate": 7e-07,
+      "loss": 0.5356,
+      "step": 5460
+    },
+    {
+      "epoch": 69.24,
+      "learning_rate": 5.333333333333333e-07,
+      "loss": 0.5289,
+      "step": 5470
+    },
+    {
+      "epoch": 69.37,
+      "learning_rate": 3.666666666666666e-07,
+      "loss": 0.5522,
+      "step": 5480
+    },
+    {
+      "epoch": 69.49,
+      "learning_rate": 1.9999999999999996e-07,
+      "loss": 0.4897,
+      "step": 5490
+    },
+    {
+      "epoch": 69.62,
+      "learning_rate": 3.3333333333333334e-08,
+      "loss": 0.5564,
+      "step": 5500
+    },
+    {
+      "epoch": 69.62,
+      "eval_cer": 0.19773618906387852,
+      "eval_loss": 0.9461079239845276,
+      "eval_runtime": 45.3275,
+      "eval_samples_per_second": 10.612,
+      "eval_steps_per_second": 1.346,
+      "eval_wer": 0.5148362892223738,
+      "step": 5500
+    },
+    {
+      "epoch": 69.75,
+      "learning_rate": 7.38e-06,
+      "loss": 0.525,
+      "step": 5510
+    },
+    {
+      "epoch": 69.87,
+      "learning_rate": 7.229999999999999e-06,
+      "loss": 0.5232,
+      "step": 5520
+    },
+    {
+      "epoch": 70.0,
+      "learning_rate": 7.079999999999999e-06,
+      "loss": 0.5318,
+      "step": 5530
+    },
+    {
+      "epoch": 70.13,
+      "learning_rate": 6.929999999999999e-06,
+      "loss": 0.562,
+      "step": 5540
+    },
+    {
+      "epoch": 70.25,
+      "learning_rate": 6.779999999999999e-06,
+      "loss": 0.494,
+      "step": 5550
+    },
+    {
+      "epoch": 70.38,
+      "learning_rate": 6.63e-06,
+      "loss": 0.5314,
+      "step": 5560
+    },
+    {
+      "epoch": 70.51,
+      "learning_rate": 6.48e-06,
+      "loss": 0.5332,
+      "step": 5570
+    },
+    {
+      "epoch": 70.63,
+      "learning_rate": 6.3299999999999995e-06,
+      "loss": 0.552,
+      "step": 5580
+    },
+    {
+      "epoch": 70.76,
+      "learning_rate": 6.179999999999999e-06,
+      "loss": 0.5538,
+      "step": 5590
+    },
+    {
+      "epoch": 70.89,
+      "learning_rate": 6.029999999999999e-06,
+      "loss": 0.5507,
+      "step": 5600
+    },
+    {
+      "epoch": 71.01,
+      "learning_rate": 5.88e-06,
+      "loss": 0.5207,
+      "step": 5610
+    },
+    {
+      "epoch": 71.14,
+      "learning_rate": 5.729999999999999e-06,
+      "loss": 0.5613,
+      "step": 5620
+    },
+    {
+      "epoch": 71.27,
+      "learning_rate": 5.579999999999999e-06,
+      "loss": 0.5263,
+      "step": 5630
+    },
+    {
+      "epoch": 71.39,
+      "learning_rate": 5.43e-06,
+      "loss": 0.5138,
+      "step": 5640
+    },
+    {
+      "epoch": 71.52,
+      "learning_rate": 5.28e-06,
+      "loss": 0.5268,
+      "step": 5650
+    },
+    {
+      "epoch": 71.65,
+      "learning_rate": 5.13e-06,
+      "loss": 0.5285,
+      "step": 5660
+    },
+    {
+      "epoch": 71.77,
+      "learning_rate": 4.98e-06,
+      "loss": 0.539,
+      "step": 5670
+    },
+    {
+      "epoch": 71.9,
+      "learning_rate": 4.8299999999999995e-06,
+      "loss": 0.5518,
+      "step": 5680
+    },
+    {
+      "epoch": 72.03,
+      "learning_rate": 4.679999999999999e-06,
+      "loss": 0.5392,
+      "step": 5690
+    },
+    {
+      "epoch": 72.15,
+      "learning_rate": 4.53e-06,
+      "loss": 0.5341,
+      "step": 5700
+    },
+    {
+      "epoch": 72.28,
+      "learning_rate": 4.3799999999999996e-06,
+      "loss": 0.528,
+      "step": 5710
+    },
+    {
+      "epoch": 72.41,
+      "learning_rate": 4.229999999999999e-06,
+      "loss": 0.5285,
+      "step": 5720
+    },
+    {
+      "epoch": 72.53,
+      "learning_rate": 4.079999999999999e-06,
+      "loss": 0.5291,
+      "step": 5730
+    },
+    {
+      "epoch": 72.66,
+      "learning_rate": 3.93e-06,
+      "loss": 0.56,
+      "step": 5740
+    },
+    {
+      "epoch": 72.78,
+      "learning_rate": 3.78e-06,
+      "loss": 0.5638,
+      "step": 5750
+    },
+    {
+      "epoch": 72.91,
+      "learning_rate": 3.6299999999999995e-06,
+      "loss": 0.5111,
+      "step": 5760
+    },
+    {
+      "epoch": 73.04,
+      "learning_rate": 3.4799999999999993e-06,
+      "loss": 0.5239,
+      "step": 5770
+    },
+    {
+      "epoch": 73.16,
+      "learning_rate": 3.33e-06,
+      "loss": 0.5386,
+      "step": 5780
+    },
+    {
+      "epoch": 73.29,
+      "learning_rate": 3.1799999999999996e-06,
+      "loss": 0.525,
+      "step": 5790
+    },
+    {
+      "epoch": 73.42,
+      "learning_rate": 3.03e-06,
+      "loss": 0.5051,
+      "step": 5800
+    },
+    {
+      "epoch": 73.54,
+      "learning_rate": 2.8799999999999995e-06,
+      "loss": 0.5119,
+      "step": 5810
+    },
+    {
+      "epoch": 73.67,
+      "learning_rate": 2.7299999999999997e-06,
+      "loss": 0.5209,
+      "step": 5820
+    },
+    {
+      "epoch": 73.8,
+      "learning_rate": 2.58e-06,
+      "loss": 0.5659,
+      "step": 5830
+    },
+    {
+      "epoch": 73.92,
+      "learning_rate": 2.4299999999999996e-06,
+      "loss": 0.5178,
+      "step": 5840
+    },
+    {
+      "epoch": 74.05,
+      "learning_rate": 2.2799999999999998e-06,
+      "loss": 0.5523,
+      "step": 5850
+    },
+    {
+      "epoch": 74.18,
+      "learning_rate": 2.13e-06,
+      "loss": 0.5048,
+      "step": 5860
+    },
+    {
+      "epoch": 74.3,
+      "learning_rate": 1.9799999999999997e-06,
+      "loss": 0.5109,
+      "step": 5870
+    },
+    {
+      "epoch": 74.43,
+      "learning_rate": 1.83e-06,
+      "loss": 0.5092,
+      "step": 5880
+    },
+    {
+      "epoch": 74.56,
+      "learning_rate": 1.6799999999999998e-06,
+      "loss": 0.5439,
+      "step": 5890
+    },
+    {
+      "epoch": 74.68,
+      "learning_rate": 1.53e-06,
+      "loss": 0.5501,
+      "step": 5900
+    },
+    {
+      "epoch": 74.81,
+      "learning_rate": 1.38e-06,
+      "loss": 0.5628,
+      "step": 5910
+    },
+    {
+      "epoch": 74.94,
+      "learning_rate": 1.23e-06,
+      "loss": 0.5097,
+      "step": 5920
+    },
+    {
+      "epoch": 75.06,
+      "learning_rate": 1.0799999999999998e-06,
+      "loss": 0.5363,
+      "step": 5930
+    },
+    {
+      "epoch": 75.19,
+      "learning_rate": 9.299999999999999e-07,
+      "loss": 0.5304,
+      "step": 5940
+    },
+    {
+      "epoch": 75.32,
+      "learning_rate": 7.799999999999999e-07,
+      "loss": 0.5358,
+      "step": 5950
+    },
+    {
+      "epoch": 75.44,
+      "learning_rate": 6.299999999999999e-07,
+      "loss": 0.5262,
+      "step": 5960
+    },
+    {
+      "epoch": 75.57,
+      "learning_rate": 4.8e-07,
+      "loss": 0.5258,
+      "step": 5970
+    },
+    {
+      "epoch": 75.7,
+      "learning_rate": 3.3e-07,
+      "loss": 0.4952,
+      "step": 5980
+    },
+    {
+      "epoch": 75.82,
+      "learning_rate": 1.7999999999999997e-07,
+      "loss": 0.5285,
+      "step": 5990
+    },
+    {
+      "epoch": 75.95,
+      "learning_rate": 3e-08,
+      "loss": 0.5252,
+      "step": 6000
+    },
+    {
+      "epoch": 75.95,
+      "eval_cer": 0.19692902596061795,
+      "eval_loss": 0.9505288004875183,
+      "eval_runtime": 45.9888,
+      "eval_samples_per_second": 10.459,
+      "eval_steps_per_second": 0.674,
+      "eval_wer": 0.5117667121418826,
+      "step": 6000
+    },
+    {
+      "epoch": 75.95,
+      "step": 6000,
+      "total_flos": 6.910110276723645e+19,
+      "train_loss": 0.044292491674423215,
+      "train_runtime": 2233.4842,
+      "train_samples_per_second": 85.964,
+      "train_steps_per_second": 2.686
+    }
+  ],
+  "max_steps": 6000,
+  "num_train_epochs": 76,
+  "total_flos": 6.910110276723645e+19,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c320c55ba9124e0a7d45da133173ad9a74af1e9959ad8e0d0bd14767c2fba239
+size 3451

vocab.json

@@ -0,0 +1,117 @@
+{
+  "[PAD]": 114,
+  "[UNK]": 113,
+  "_": 1,
+  "`": 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,
+  "|": 0,
+  "¥": 29,
+  "°": 30,
+  "½": 31,
+  "¾": 32,
+  "é": 33,
+  "í": 34,
+  "،": 35,
+  "؛": 36,
+  "؟": 37,
+  "ء": 38,
+  "آ": 39,
+  "أ": 40,
+  "ؤ": 41,
+  "ئ": 42,
+  "ا": 43,
+  "ب": 44,
+  "ت": 45,
+  "ث": 46,
+  "ج": 47,
+  "ح": 48,
+  "خ": 49,
+  "د": 50,
+  "ذ": 51,
+  "ر": 52,
+  "ز": 53,
+  "س": 54,
+  "ش": 55,
+  "ص": 56,
+  "ض": 57,
+  "ط": 58,
+  "ظ": 59,
+  "ع": 60,
+  "غ": 61,
+  "ـ": 62,
+  "ف": 63,
+  "ق": 64,
+  "ك": 65,
+  "ل": 66,
+  "م": 67,
+  "ن": 68,
+  "ه": 69,
+  "و": 70,
+  "ى": 71,
+  "ي": 72,
+  "ً": 73,
+  "ٌ": 74,
+  "َ": 75,
+  "ُ": 76,
+  "ّ": 77,
+  "٪": 78,
+  "ټ": 79,
+  "پ": 80,
+  "ځ": 81,
+  "څ": 82,
+  "چ": 83,
+  "ډ": 84,
+  "ړ": 85,
+  "ږ": 86,
+  "ژ": 87,
+  "ښ": 88,
+  "ک": 89,
+  "ګ": 90,
+  "گ": 91,
+  "ڼ": 92,
+  "ھ": 93,
+  "ی": 94,
+  "ۍ": 95,
+  "ې": 96,
+  "ے": 97,
+  "۔": 98,
+  "۰": 99,
+  "۱": 100,
+  "۲": 101,
+  "۳": 102,
+  "۴": 103,
+  "۵": 104,
+  "۶": 105,
+  "۷": 106,
+  "۸": 107,
+  "۹": 108,
+  "‌": 109,
+  "‍": 110,
+  "–": 111,
+  "—": 112
+}