add model without lm

README.md

@@ -0,0 +1,119 @@
+---
+tags:
+- automatic-speech-recognition
+- Lemswasabi/tuudle
+- generated_from_trainer
+datasets:
+- tuudle
+model-index:
+- name: ''
+  results: []
+---
+
+<!-- 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. -->
+
+# 
+
+This model is a fine-tuned version of [Lemswasabi/letzspeak](https://huggingface.co/Lemswasabi/letzspeak) on the LEMSWASABI/TUUDLE - RTL dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.1058
+- Wer: 0.1075
+
+## 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-05
+- train_batch_size: 3
+- eval_batch_size: 3
+- seed: 42
+- gradient_accumulation_steps: 4
+- total_train_batch_size: 12
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- lr_scheduler_warmup_steps: 2000
+- num_epochs: 50.0
+- mixed_precision_training: Native AMP
+
+### Training results
+
+| Training Loss | Epoch | Step  | Validation Loss | Wer    |
+|:-------------:|:-----:|:-----:|:---------------:|:------:|
+| 3.1484        | 0.89  | 500   | 3.0844          | 1.0    |
+| 2.6539        | 1.77  | 1000  | 1.7272          | 0.9358 |
+| 0.8732        | 2.66  | 1500  | 0.1975          | 0.1609 |
+| 0.8075        | 3.55  | 2000  | 0.1483          | 0.1468 |
+| 0.7358        | 4.43  | 2500  | 0.1331          | 0.1401 |
+| 0.7079        | 5.32  | 3000  | 0.1273          | 0.1364 |
+| 0.7032        | 6.21  | 3500  | 0.1133          | 0.1240 |
+| 0.7129        | 7.09  | 4000  | 0.1124          | 0.1290 |
+| 0.6771        | 7.98  | 4500  | 0.1121          | 0.1300 |
+| 0.6859        | 8.86  | 5000  | 0.1095          | 0.1313 |
+| 0.6496        | 9.75  | 5500  | 0.1091          | 0.1250 |
+| 0.6431        | 10.64 | 6000  | 0.1102          | 0.1293 |
+| 0.6422        | 11.52 | 6500  | 0.1107          | 0.1179 |
+| 0.6334        | 12.41 | 7000  | 0.1049          | 0.1236 |
+| 0.599         | 13.3  | 7500  | 0.1092          | 0.1152 |
+| 0.6205        | 14.18 | 8000  | 0.1047          | 0.1219 |
+| 0.5944        | 15.07 | 8500  | 0.1068          | 0.1203 |
+| 0.6102        | 15.96 | 9000  | 0.1056          | 0.1159 |
+| 0.5983        | 16.84 | 9500  | 0.1061          | 0.1152 |
+| 0.5882        | 17.73 | 10000 | 0.1043          | 0.1135 |
+| 0.5876        | 18.62 | 10500 | 0.1023          | 0.1159 |
+| 0.5717        | 19.5  | 11000 | 0.1037          | 0.1233 |
+| 0.5537        | 20.39 | 11500 | 0.1070          | 0.1192 |
+| 0.5636        | 21.28 | 12000 | 0.1036          | 0.1169 |
+| 0.5536        | 22.16 | 12500 | 0.1008          | 0.1182 |
+| 0.5656        | 23.05 | 13000 | 0.1010          | 0.1172 |
+| 0.5504        | 23.94 | 13500 | 0.1019          | 0.1105 |
+| 0.5476        | 24.82 | 14000 | 0.1026          | 0.1166 |
+| 0.5375        | 25.71 | 14500 | 0.1107          | 0.1189 |
+| 0.5318        | 26.6  | 15000 | 0.1051          | 0.1142 |
+| 0.5278        | 27.48 | 15500 | 0.1049          | 0.1166 |
+| 0.5204        | 28.37 | 16000 | 0.1081          | 0.1182 |
+| 0.512         | 29.26 | 16500 | 0.1062          | 0.1156 |
+| 0.5082        | 30.14 | 17000 | 0.1045          | 0.1135 |
+| 0.5193        | 31.03 | 17500 | 0.1091          | 0.1145 |
+| 0.5129        | 31.91 | 18000 | 0.1040          | 0.1088 |
+| 0.5126        | 32.8  | 18500 | 0.1085          | 0.1169 |
+| 0.496         | 33.69 | 19000 | 0.1070          | 0.1166 |
+| 0.5017        | 34.57 | 19500 | 0.1119          | 0.1162 |
+| 0.4808        | 35.46 | 20000 | 0.1101          | 0.1139 |
+| 0.4939        | 36.35 | 20500 | 0.1081          | 0.1125 |
+| 0.4738        | 37.23 | 21000 | 0.1091          | 0.1098 |
+| 0.4978        | 38.12 | 21500 | 0.1057          | 0.1092 |
+| 0.4972        | 39.01 | 22000 | 0.1074          | 0.1105 |
+| 0.4773        | 39.89 | 22500 | 0.1062          | 0.1108 |
+| 0.4741        | 40.78 | 23000 | 0.1057          | 0.1085 |
+| 0.4776        | 41.67 | 23500 | 0.1077          | 0.1085 |
+| 0.4637        | 42.55 | 24000 | 0.1061          | 0.1095 |
+| 0.4853        | 43.44 | 24500 | 0.1081          | 0.1075 |
+| 0.4602        | 44.33 | 25000 | 0.1076          | 0.1085 |
+| 0.4667        | 45.21 | 25500 | 0.1078          | 0.1078 |
+| 0.4484        | 46.1  | 26000 | 0.1056          | 0.1082 |
+| 0.4601        | 46.99 | 26500 | 0.1066          | 0.1078 |
+| 0.4691        | 47.87 | 27000 | 0.1068          | 0.1085 |
+| 0.4457        | 48.76 | 27500 | 0.1066          | 0.1078 |
+| 0.475         | 49.65 | 28000 | 0.1060          | 0.1082 |
+
+
+### Framework versions
+
+- Transformers 4.20.0.dev0
+- Pytorch 1.11.0+cu113
+- Datasets 2.2.1
+- Tokenizers 0.12.1

added_tokens.json

@@ -0,0 +1 @@
+{"<s>": 57, "</s>": 58}

all_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 50.0,
+    "eval_loss": 0.10580451041460037,
+    "eval_runtime": 31.0996,
+    "eval_samples": 178,
+    "eval_samples_per_second": 5.724,
+    "eval_steps_per_second": 1.929,
+    "eval_wer": 0.10749076251259658,
+    "train_loss": 0.6889870901987062,
+    "train_runtime": 132713.9902,
+    "train_samples": 6770,
+    "train_samples_per_second": 2.551,
+    "train_steps_per_second": 0.212
+}

config.json

@@ -0,0 +1,115 @@
+{
+  "_name_or_path": "Lemswasabi/letzspeak",
+  "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_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.25,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.75,
+  "mask_time_selection": "static",
+  "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": 56,
+  "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": "4.20.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 59,
+  "xvector_output_dim": 512
+}

eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 50.0,
+    "eval_loss": 0.10580451041460037,
+    "eval_runtime": 31.0996,
+    "eval_samples": 178,
+    "eval_samples_per_second": 5.724,
+    "eval_steps_per_second": 1.929,
+    "eval_wer": 0.10749076251259658
+}

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:d35b9085ec4fe5c026ae1207feb043e4df0016d99055a3a68d93bf0255c37980
+size 1262140593

run.sh

@@ -0,0 +1,37 @@
+python run_speech_recognition_ctc.py \
+	--dataset_name="Lemswasabi/tuudle" \
+	--model_name_or_path="Lemswasabi/letzspeak" \
+	--dataset_config_name="rtl" \
+	--train_split_name="train" \
+	--eval_split_name="validation" \
+	--output_dir="./" \
+	--overwrite_output_dir \
+	--num_train_epochs="50" \
+	--per_device_train_batch_size="3" \
+	--per_device_eval_batch_size="3" \
+	--gradient_accumulation_steps="4" \
+	--learning_rate="7.5e-5" \
+	--warmup_steps="2000" \
+	--length_column_name="input_length" \
+	--evaluation_strategy="steps" \
+	--text_column_name="sentence" \
+	--chars_to_replace \- \
+	--chars_to_ignore , ? . ! \; \: \" “ % ‘ „ ” � — ’ … – \
+	--save_steps="500" \
+	--eval_steps="500" \
+	--logging_steps="100" \
+	--layerdrop="0.0" \
+	--activation_dropout="0.1" \
+	--save_total_limit="3" \
+	--freeze_feature_encoder \
+	--feat_proj_dropout="0.0" \
+	--mask_time_prob="0.75" \
+	--mask_time_length="10" \
+	--mask_feature_prob="0.25" \
+	--mask_feature_length="64" \
+	--gradient_checkpointing \
+	--use_auth_token \
+	--fp16 \
+	--group_by_length \
+	--do_train --do_eval \
+	--push_to_hub

run_speech_recognition_ctc.py

@@ -0,0 +1,780 @@
+#!/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, load_metric
+
+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
+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("4.20.0.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."},
+    )
+    chars_to_replace: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to replace 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:`transformers-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",
+        )
+
+        with self.processor.as_target_processor():
+            labels_batch = self.processor.pad(
+                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
+
+        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()
+
+    # 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_replace_regex = (
+        f'[{"".join(data_args.chars_to_replace)}]' if data_args.chars_to_replace is not None else None
+    )
+    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 replace_special_characters(batch):
+        if chars_to_replace_regex is not None:
+            batch[text_column_name] = re.sub(chars_to_replace_regex, " ", batch[text_column_name]).lower() + " "
+        else:
+            batch[text_column_name] = batch[text_column_name].lower() + " "
+        return batch
+
+    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(
+            replace_special_characters,
+            desc="replace special characters from datasets",
+        )
+        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):
+                os.remove(vocab_file)
+
+        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,
+        ignore_mismatched_sizes=True,
+    )
+
+    # 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: load_metric(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": "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/May21_09-02-48_bioman2/1653117325.7043028/events.out.tfevents.1653117325.bioman2.19214.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82fb10ff9df3efb104856e487ef603c660eb2f153aa260fd17fbe81016b0af6c
+size 5120

runs/May21_09-02-48_bioman2/events.out.tfevents.1653117325.bioman2.19214.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a7c807f6a57e6444980141ea9ca5f0c2b26ea07cfe67afcbc2a3bb0ec031e0b
+size 68029

runs/May21_09-02-48_bioman2/events.out.tfevents.1653250459.bioman2.19214.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e0d53e20940f64d80f7ff85ddefa514142b54a7fec3fbf3af653b1603f5407c
+size 364

special_tokens_map.json

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

tokenizer_config.json

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

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 50.0,
+    "train_loss": 0.6889870901987062,
+    "train_runtime": 132713.9902,
+    "train_samples": 6770,
+    "train_samples_per_second": 2.551,
+    "train_steps_per_second": 0.212
+}

trainer_state.json

@@ -0,0 +1,2221 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 49.999556933983165,
+  "global_step": 28200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.7125e-06,
+      "loss": 9.6392,
+      "step": 100
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 7.4625e-06,
+      "loss": 5.3472,
+      "step": 200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.1212499999999998e-05,
+      "loss": 3.7052,
+      "step": 300
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.49625e-05,
+      "loss": 3.35,
+      "step": 400
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.8712499999999997e-05,
+      "loss": 3.1484,
+      "step": 500
+    },
+    {
+      "epoch": 0.89,
+      "eval_loss": 3.084352731704712,
+      "eval_runtime": 31.0524,
+      "eval_samples_per_second": 5.732,
+      "eval_steps_per_second": 1.932,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 2.2462499999999997e-05,
+      "loss": 3.0531,
+      "step": 600
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 2.6212499999999997e-05,
+      "loss": 2.9911,
+      "step": 700
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 2.99625e-05,
+      "loss": 2.9576,
+      "step": 800
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 3.37125e-05,
+      "loss": 2.9406,
+      "step": 900
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 3.7462499999999996e-05,
+      "loss": 2.6539,
+      "step": 1000
+    },
+    {
+      "epoch": 1.77,
+      "eval_loss": 1.7271525859832764,
+      "eval_runtime": 30.9027,
+      "eval_samples_per_second": 5.76,
+      "eval_steps_per_second": 1.942,
+      "eval_wer": 0.9358414511252939,
+      "step": 1000
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 4.12125e-05,
+      "loss": 1.5429,
+      "step": 1100
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 4.4962499999999995e-05,
+      "loss": 1.0875,
+      "step": 1200
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 4.871249999999999e-05,
+      "loss": 0.9728,
+      "step": 1300
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 5.2424999999999994e-05,
+      "loss": 0.9117,
+      "step": 1400
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 5.6175e-05,
+      "loss": 0.8732,
+      "step": 1500
+    },
+    {
+      "epoch": 2.66,
+      "eval_loss": 0.19752363860607147,
+      "eval_runtime": 31.0609,
+      "eval_samples_per_second": 5.731,
+      "eval_steps_per_second": 1.932,
+      "eval_wer": 0.16090023513604298,
+      "step": 1500
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 5.9925e-05,
+      "loss": 0.8518,
+      "step": 1600
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 6.367499999999999e-05,
+      "loss": 0.8373,
+      "step": 1700
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 6.7425e-05,
+      "loss": 0.7958,
+      "step": 1800
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 7.1175e-05,
+      "loss": 0.8058,
+      "step": 1900
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 7.492499999999999e-05,
+      "loss": 0.8075,
+      "step": 2000
+    },
+    {
+      "epoch": 3.55,
+      "eval_loss": 0.1483038365840912,
+      "eval_runtime": 31.0609,
+      "eval_samples_per_second": 5.731,
+      "eval_steps_per_second": 1.932,
+      "eval_wer": 0.1467920725562647,
+      "step": 2000
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 7.471946564885497e-05,
+      "loss": 0.785,
+      "step": 2100
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 7.443320610687022e-05,
+      "loss": 0.7656,
+      "step": 2200
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 7.414694656488549e-05,
+      "loss": 0.7681,
+      "step": 2300
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 7.386068702290076e-05,
+      "loss": 0.7535,
+      "step": 2400
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 7.357442748091603e-05,
+      "loss": 0.7358,
+      "step": 2500
+    },
+    {
+      "epoch": 4.43,
+      "eval_loss": 0.1331482231616974,
+      "eval_runtime": 30.9431,
+      "eval_samples_per_second": 5.752,
+      "eval_steps_per_second": 1.939,
+      "eval_wer": 0.14007389989922742,
+      "step": 2500
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 7.328816793893128e-05,
+      "loss": 0.7478,
+      "step": 2600
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 7.300190839694656e-05,
+      "loss": 0.746,
+      "step": 2700
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 7.271564885496182e-05,
+      "loss": 0.7545,
+      "step": 2800
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 7.242938931297709e-05,
+      "loss": 0.7268,
+      "step": 2900
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 7.214312977099236e-05,
+      "loss": 0.7079,
+      "step": 3000
+    },
+    {
+      "epoch": 5.32,
+      "eval_loss": 0.12731623649597168,
+      "eval_runtime": 30.9985,
+      "eval_samples_per_second": 5.742,
+      "eval_steps_per_second": 1.936,
+      "eval_wer": 0.1363789049378569,
+      "step": 3000
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 7.185687022900763e-05,
+      "loss": 0.7344,
+      "step": 3100
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 7.15706106870229e-05,
+      "loss": 0.7334,
+      "step": 3200
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 7.128435114503815e-05,
+      "loss": 0.7274,
+      "step": 3300
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 7.099809160305343e-05,
+      "loss": 0.7271,
+      "step": 3400
+    },
+    {
+      "epoch": 6.21,
+      "learning_rate": 7.071469465648854e-05,
+      "loss": 0.7032,
+      "step": 3500
+    },
+    {
+      "epoch": 6.21,
+      "eval_loss": 0.11333052814006805,
+      "eval_runtime": 31.1935,
+      "eval_samples_per_second": 5.706,
+      "eval_steps_per_second": 1.923,
+      "eval_wer": 0.12395028552233793,
+      "step": 3500
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 7.04284351145038e-05,
+      "loss": 0.6852,
+      "step": 3600
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 7.014217557251907e-05,
+      "loss": 0.699,
+      "step": 3700
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 6.985591603053435e-05,
+      "loss": 0.6894,
+      "step": 3800
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 6.956965648854962e-05,
+      "loss": 0.7014,
+      "step": 3900
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 6.928339694656487e-05,
+      "loss": 0.7129,
+      "step": 4000
+    },
+    {
+      "epoch": 7.09,
+      "eval_loss": 0.11242285370826721,
+      "eval_runtime": 31.0672,
+      "eval_samples_per_second": 5.73,
+      "eval_steps_per_second": 1.931,
+      "eval_wer": 0.1289889150151159,
+      "step": 4000
+    },
+    {
+      "epoch": 7.27,
+      "learning_rate": 6.899713740458015e-05,
+      "loss": 0.6902,
+      "step": 4100
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 6.871087786259541e-05,
+      "loss": 0.6898,
+      "step": 4200
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 6.842461832061069e-05,
+      "loss": 0.6785,
+      "step": 4300
+    },
+    {
+      "epoch": 7.8,
+      "learning_rate": 6.813835877862594e-05,
+      "loss": 0.7024,
+      "step": 4400
+    },
+    {
+      "epoch": 7.98,
+      "learning_rate": 6.785209923664122e-05,
+      "loss": 0.6771,
+      "step": 4500
+    },
+    {
+      "epoch": 7.98,
+      "eval_loss": 0.11212227493524551,
+      "eval_runtime": 31.0691,
+      "eval_samples_per_second": 5.729,
+      "eval_steps_per_second": 1.931,
+      "eval_wer": 0.12999664091367147,
+      "step": 4500
+    },
+    {
+      "epoch": 8.16,
+      "learning_rate": 6.756583969465648e-05,
+      "loss": 0.6555,
+      "step": 4600
+    },
+    {
+      "epoch": 8.33,
+      "learning_rate": 6.727958015267174e-05,
+      "loss": 0.6897,
+      "step": 4700
+    },
+    {
+      "epoch": 8.51,
+      "learning_rate": 6.699618320610687e-05,
+      "loss": 0.6809,
+      "step": 4800
+    },
+    {
+      "epoch": 8.69,
+      "learning_rate": 6.670992366412213e-05,
+      "loss": 0.6821,
+      "step": 4900
+    },
+    {
+      "epoch": 8.86,
+      "learning_rate": 6.642366412213741e-05,
+      "loss": 0.6859,
+      "step": 5000
+    },
+    {
+      "epoch": 8.86,
+      "eval_loss": 0.109534852206707,
+      "eval_runtime": 31.3176,
+      "eval_samples_per_second": 5.684,
+      "eval_steps_per_second": 1.916,
+      "eval_wer": 0.13134027544507895,
+      "step": 5000
+    },
+    {
+      "epoch": 9.04,
+      "learning_rate": 6.613740458015266e-05,
+      "loss": 0.6849,
+      "step": 5100
+    },
+    {
+      "epoch": 9.22,
+      "learning_rate": 6.585114503816793e-05,
+      "loss": 0.6615,
+      "step": 5200
+    },
+    {
+      "epoch": 9.4,
+      "learning_rate": 6.55648854961832e-05,
+      "loss": 0.6587,
+      "step": 5300
+    },
+    {
+      "epoch": 9.57,
+      "learning_rate": 6.527862595419846e-05,
+      "loss": 0.6741,
+      "step": 5400
+    },
+    {
+      "epoch": 9.75,
+      "learning_rate": 6.499236641221373e-05,
+      "loss": 0.6496,
+      "step": 5500
+    },
+    {
+      "epoch": 9.75,
+      "eval_loss": 0.1090548112988472,
+      "eval_runtime": 31.0434,
+      "eval_samples_per_second": 5.734,
+      "eval_steps_per_second": 1.933,
+      "eval_wer": 0.12495801142089352,
+      "step": 5500
+    },
+    {
+      "epoch": 9.93,
+      "learning_rate": 6.4706106870229e-05,
+      "loss": 0.6672,
+      "step": 5600
+    },
+    {
+      "epoch": 10.11,
+      "learning_rate": 6.441984732824428e-05,
+      "loss": 0.6511,
+      "step": 5700
+    },
+    {
+      "epoch": 10.28,
+      "learning_rate": 6.413645038167938e-05,
+      "loss": 0.6393,
+      "step": 5800
+    },
+    {
+      "epoch": 10.46,
+      "learning_rate": 6.385019083969465e-05,
+      "loss": 0.6531,
+      "step": 5900
+    },
+    {
+      "epoch": 10.64,
+      "learning_rate": 6.356393129770992e-05,
+      "loss": 0.6431,
+      "step": 6000
+    },
+    {
+      "epoch": 10.64,
+      "eval_loss": 0.11021342128515244,
+      "eval_runtime": 30.9854,
+      "eval_samples_per_second": 5.745,
+      "eval_steps_per_second": 1.936,
+      "eval_wer": 0.12932482364796774,
+      "step": 6000
+    },
+    {
+      "epoch": 10.82,
+      "learning_rate": 6.32776717557252e-05,
+      "loss": 0.6476,
+      "step": 6100
+    },
+    {
+      "epoch": 10.99,
+      "learning_rate": 6.299141221374044e-05,
+      "loss": 0.6504,
+      "step": 6200
+    },
+    {
+      "epoch": 11.17,
+      "learning_rate": 6.270515267175572e-05,
+      "loss": 0.6366,
+      "step": 6300
+    },
+    {
+      "epoch": 11.35,
+      "learning_rate": 6.241889312977098e-05,
+      "loss": 0.6435,
+      "step": 6400
+    },
+    {
+      "epoch": 11.52,
+      "learning_rate": 6.213263358778625e-05,
+      "loss": 0.6422,
+      "step": 6500
+    },
+    {
+      "epoch": 11.52,
+      "eval_loss": 0.11072035878896713,
+      "eval_runtime": 31.126,
+      "eval_samples_per_second": 5.719,
+      "eval_steps_per_second": 1.928,
+      "eval_wer": 0.11790393013100436,
+      "step": 6500
+    },
+    {
+      "epoch": 11.7,
+      "learning_rate": 6.184637404580152e-05,
+      "loss": 0.636,
+      "step": 6600
+    },
+    {
+      "epoch": 11.88,
+      "learning_rate": 6.156011450381679e-05,
+      "loss": 0.647,
+      "step": 6700
+    },
+    {
+      "epoch": 12.06,
+      "learning_rate": 6.127385496183207e-05,
+      "loss": 0.644,
+      "step": 6800
+    },
+    {
+      "epoch": 12.23,
+      "learning_rate": 6.098759541984732e-05,
+      "loss": 0.611,
+      "step": 6900
+    },
+    {
+      "epoch": 12.41,
+      "learning_rate": 6.070133587786259e-05,
+      "loss": 0.6334,
+      "step": 7000
+    },
+    {
+      "epoch": 12.41,
+      "eval_loss": 0.10494451969861984,
+      "eval_runtime": 31.1005,
+      "eval_samples_per_second": 5.723,
+      "eval_steps_per_second": 1.929,
+      "eval_wer": 0.12361437688948607,
+      "step": 7000
+    },
+    {
+      "epoch": 12.59,
+      "learning_rate": 6.0415076335877855e-05,
+      "loss": 0.6305,
+      "step": 7100
+    },
+    {
+      "epoch": 12.77,
+      "learning_rate": 6.0128816793893126e-05,
+      "loss": 0.6292,
+      "step": 7200
+    },
+    {
+      "epoch": 12.94,
+      "learning_rate": 5.984255725190839e-05,
+      "loss": 0.6341,
+      "step": 7300
+    },
+    {
+      "epoch": 13.12,
+      "learning_rate": 5.955629770992366e-05,
+      "loss": 0.625,
+      "step": 7400
+    },
+    {
+      "epoch": 13.3,
+      "learning_rate": 5.927003816793893e-05,
+      "loss": 0.599,
+      "step": 7500
+    },
+    {
+      "epoch": 13.3,
+      "eval_loss": 0.10921534150838852,
+      "eval_runtime": 31.0896,
+      "eval_samples_per_second": 5.725,
+      "eval_steps_per_second": 1.93,
+      "eval_wer": 0.11521666106818945,
+      "step": 7500
+    },
+    {
+      "epoch": 13.47,
+      "learning_rate": 5.898377862595419e-05,
+      "loss": 0.6144,
+      "step": 7600
+    },
+    {
+      "epoch": 13.65,
+      "learning_rate": 5.869751908396946e-05,
+      "loss": 0.6314,
+      "step": 7700
+    },
+    {
+      "epoch": 13.83,
+      "learning_rate": 5.8411259541984726e-05,
+      "loss": 0.6185,
+      "step": 7800
+    },
+    {
+      "epoch": 14.01,
+      "learning_rate": 5.8124999999999997e-05,
+      "loss": 0.6122,
+      "step": 7900
+    },
+    {
+      "epoch": 14.18,
+      "learning_rate": 5.783874045801526e-05,
+      "loss": 0.6205,
+      "step": 8000
+    },
+    {
+      "epoch": 14.18,
+      "eval_loss": 0.1046827957034111,
+      "eval_runtime": 31.2229,
+      "eval_samples_per_second": 5.701,
+      "eval_steps_per_second": 1.922,
+      "eval_wer": 0.12193483372522673,
+      "step": 8000
+    },
+    {
+      "epoch": 14.36,
+      "learning_rate": 5.755248091603053e-05,
+      "loss": 0.612,
+      "step": 8100
+    },
+    {
+      "epoch": 14.54,
+      "learning_rate": 5.7266221374045794e-05,
+      "loss": 0.6196,
+      "step": 8200
+    },
+    {
+      "epoch": 14.72,
+      "learning_rate": 5.6979961832061064e-05,
+      "loss": 0.6114,
+      "step": 8300
+    },
+    {
+      "epoch": 14.89,
+      "learning_rate": 5.6693702290076334e-05,
+      "loss": 0.6005,
+      "step": 8400
+    },
+    {
+      "epoch": 15.07,
+      "learning_rate": 5.64074427480916e-05,
+      "loss": 0.5944,
+      "step": 8500
+    },
+    {
+      "epoch": 15.07,
+      "eval_loss": 0.10684490948915482,
+      "eval_runtime": 31.2028,
+      "eval_samples_per_second": 5.705,
+      "eval_steps_per_second": 1.923,
+      "eval_wer": 0.12025529056096741,
+      "step": 8500
+    },
+    {
+      "epoch": 15.25,
+      "learning_rate": 5.612118320610687e-05,
+      "loss": 0.6108,
+      "step": 8600
+    },
+    {
+      "epoch": 15.43,
+      "learning_rate": 5.583492366412213e-05,
+      "loss": 0.5995,
+      "step": 8700
+    },
+    {
+      "epoch": 15.6,
+      "learning_rate": 5.55486641221374e-05,
+      "loss": 0.6117,
+      "step": 8800
+    },
+    {
+      "epoch": 15.78,
+      "learning_rate": 5.5262404580152665e-05,
+      "loss": 0.6006,
+      "step": 8900
+    },
+    {
+      "epoch": 15.96,
+      "learning_rate": 5.4976145038167935e-05,
+      "loss": 0.6102,
+      "step": 9000
+    },
+    {
+      "epoch": 15.96,
+      "eval_loss": 0.10560546070337296,
+      "eval_runtime": 31.2837,
+      "eval_samples_per_second": 5.69,
+      "eval_steps_per_second": 1.918,
+      "eval_wer": 0.11588847833389318,
+      "step": 9000
+    },
+    {
+      "epoch": 16.13,
+      "learning_rate": 5.46898854961832e-05,
+      "loss": 0.5968,
+      "step": 9100
+    },
+    {
+      "epoch": 16.31,
+      "learning_rate": 5.440362595419847e-05,
+      "loss": 0.5912,
+      "step": 9200
+    },
+    {
+      "epoch": 16.49,
+      "learning_rate": 5.411736641221374e-05,
+      "loss": 0.5889,
+      "step": 9300
+    },
+    {
+      "epoch": 16.67,
+      "learning_rate": 5.3831106870229e-05,
+      "loss": 0.6014,
+      "step": 9400
+    },
+    {
+      "epoch": 16.84,
+      "learning_rate": 5.354484732824427e-05,
+      "loss": 0.5983,
+      "step": 9500
+    },
+    {
+      "epoch": 16.84,
+      "eval_loss": 0.1061149537563324,
+      "eval_runtime": 31.4987,
+      "eval_samples_per_second": 5.651,
+      "eval_steps_per_second": 1.905,
+      "eval_wer": 0.11521666106818945,
+      "step": 9500
+    },
+    {
+      "epoch": 17.02,
+      "learning_rate": 5.3258587786259536e-05,
+      "loss": 0.5992,
+      "step": 9600
+    },
+    {
+      "epoch": 17.2,
+      "learning_rate": 5.2972328244274806e-05,
+      "loss": 0.5887,
+      "step": 9700
+    },
+    {
+      "epoch": 17.38,
+      "learning_rate": 5.268606870229007e-05,
+      "loss": 0.6012,
+      "step": 9800
+    },
+    {
+      "epoch": 17.55,
+      "learning_rate": 5.239980916030534e-05,
+      "loss": 0.5985,
+      "step": 9900
+    },
+    {
+      "epoch": 17.73,
+      "learning_rate": 5.211354961832061e-05,
+      "loss": 0.5882,
+      "step": 10000
+    },
+    {
+      "epoch": 17.73,
+      "eval_loss": 0.10430345684289932,
+      "eval_runtime": 30.988,
+      "eval_samples_per_second": 5.744,
+      "eval_steps_per_second": 1.936,
+      "eval_wer": 0.11353711790393013,
+      "step": 10000
+    },
+    {
+      "epoch": 17.91,
+      "learning_rate": 5.182729007633587e-05,
+      "loss": 0.5906,
+      "step": 10100
+    },
+    {
+      "epoch": 18.09,
+      "learning_rate": 5.1541030534351143e-05,
+      "loss": 0.5843,
+      "step": 10200
+    },
+    {
+      "epoch": 18.26,
+      "learning_rate": 5.1257633587786254e-05,
+      "loss": 0.5976,
+      "step": 10300
+    },
+    {
+      "epoch": 18.44,
+      "learning_rate": 5.0971374045801525e-05,
+      "loss": 0.59,
+      "step": 10400
+    },
+    {
+      "epoch": 18.62,
+      "learning_rate": 5.068511450381679e-05,
+      "loss": 0.5876,
+      "step": 10500
+    },
+    {
+      "epoch": 18.62,
+      "eval_loss": 0.10231117904186249,
+      "eval_runtime": 31.0041,
+      "eval_samples_per_second": 5.741,
+      "eval_steps_per_second": 1.935,
+      "eval_wer": 0.11588847833389318,
+      "step": 10500
+    },
+    {
+      "epoch": 18.79,
+      "learning_rate": 5.039885496183206e-05,
+      "loss": 0.5845,
+      "step": 10600
+    },
+    {
+      "epoch": 18.97,
+      "learning_rate": 5.011259541984732e-05,
+      "loss": 0.5812,
+      "step": 10700
+    },
+    {
+      "epoch": 19.15,
+      "learning_rate": 4.982633587786259e-05,
+      "loss": 0.5695,
+      "step": 10800
+    },
+    {
+      "epoch": 19.33,
+      "learning_rate": 4.954007633587786e-05,
+      "loss": 0.5665,
+      "step": 10900
+    },
+    {
+      "epoch": 19.5,
+      "learning_rate": 4.9253816793893125e-05,
+      "loss": 0.5717,
+      "step": 11000
+    },
+    {
+      "epoch": 19.5,
+      "eval_loss": 0.10367337614297867,
+      "eval_runtime": 31.2366,
+      "eval_samples_per_second": 5.698,
+      "eval_steps_per_second": 1.921,
+      "eval_wer": 0.1232784682566342,
+      "step": 11000
+    },
+    {
+      "epoch": 19.68,
+      "learning_rate": 4.8967557251908396e-05,
+      "loss": 0.5657,
+      "step": 11100
+    },
+    {
+      "epoch": 19.86,
+      "learning_rate": 4.868129770992366e-05,
+      "loss": 0.5733,
+      "step": 11200
+    },
+    {
+      "epoch": 20.04,
+      "learning_rate": 4.839503816793893e-05,
+      "loss": 0.5816,
+      "step": 11300
+    },
+    {
+      "epoch": 20.21,
+      "learning_rate": 4.810877862595419e-05,
+      "loss": 0.5694,
+      "step": 11400
+    },
+    {
+      "epoch": 20.39,
+      "learning_rate": 4.782251908396946e-05,
+      "loss": 0.5537,
+      "step": 11500
+    },
+    {
+      "epoch": 20.39,
+      "eval_loss": 0.10704291611909866,
+      "eval_runtime": 31.0837,
+      "eval_samples_per_second": 5.726,
+      "eval_steps_per_second": 1.93,
+      "eval_wer": 0.11924756466241182,
+      "step": 11500
+    },
+    {
+      "epoch": 20.57,
+      "learning_rate": 4.7536259541984726e-05,
+      "loss": 0.5726,
+      "step": 11600
+    },
+    {
+      "epoch": 20.74,
+      "learning_rate": 4.7249999999999997e-05,
+      "loss": 0.5742,
+      "step": 11700
+    },
+    {
+      "epoch": 20.92,
+      "learning_rate": 4.696374045801527e-05,
+      "loss": 0.563,
+      "step": 11800
+    },
+    {
+      "epoch": 21.1,
+      "learning_rate": 4.667748091603053e-05,
+      "loss": 0.5655,
+      "step": 11900
+    },
+    {
+      "epoch": 21.28,
+      "learning_rate": 4.63912213740458e-05,
+      "loss": 0.5636,
+      "step": 12000
+    },
+    {
+      "epoch": 21.28,
+      "eval_loss": 0.10363561660051346,
+      "eval_runtime": 31.1987,
+      "eval_samples_per_second": 5.705,
+      "eval_steps_per_second": 1.923,
+      "eval_wer": 0.11689620423244877,
+      "step": 12000
+    },
+    {
+      "epoch": 21.45,
+      "learning_rate": 4.6104961832061064e-05,
+      "loss": 0.5612,
+      "step": 12100
+    },
+    {
+      "epoch": 21.63,
+      "learning_rate": 4.5818702290076334e-05,
+      "loss": 0.5732,
+      "step": 12200
+    },
+    {
+      "epoch": 21.81,
+      "learning_rate": 4.55324427480916e-05,
+      "loss": 0.5726,
+      "step": 12300
+    },
+    {
+      "epoch": 21.99,
+      "learning_rate": 4.524618320610687e-05,
+      "loss": 0.5547,
+      "step": 12400
+    },
+    {
+      "epoch": 22.16,
+      "learning_rate": 4.495992366412213e-05,
+      "loss": 0.5536,
+      "step": 12500
+    },
+    {
+      "epoch": 22.16,
+      "eval_loss": 0.10082551091909409,
+      "eval_runtime": 31.121,
+      "eval_samples_per_second": 5.72,
+      "eval_steps_per_second": 1.928,
+      "eval_wer": 0.11823983876385623,
+      "step": 12500
+    },
+    {
+      "epoch": 22.34,
+      "learning_rate": 4.46736641221374e-05,
+      "loss": 0.5509,
+      "step": 12600
+    },
+    {
+      "epoch": 22.52,
+      "learning_rate": 4.439026717557252e-05,
+      "loss": 0.5645,
+      "step": 12700
+    },
+    {
+      "epoch": 22.69,
+      "learning_rate": 4.410400763358778e-05,
+      "loss": 0.5527,
+      "step": 12800
+    },
+    {
+      "epoch": 22.87,
+      "learning_rate": 4.381774809160305e-05,
+      "loss": 0.547,
+      "step": 12900
+    },
+    {
+      "epoch": 23.05,
+      "learning_rate": 4.3534351145038163e-05,
+      "loss": 0.5656,
+      "step": 13000
+    },
+    {
+      "epoch": 23.05,
+      "eval_loss": 0.10101909935474396,
+      "eval_runtime": 30.9319,
+      "eval_samples_per_second": 5.755,
+      "eval_steps_per_second": 1.94,
+      "eval_wer": 0.11723211286530064,
+      "step": 13000
+    },
+    {
+      "epoch": 23.23,
+      "learning_rate": 4.3248091603053434e-05,
+      "loss": 0.5483,
+      "step": 13100
+    },
+    {
+      "epoch": 23.4,
+      "learning_rate": 4.29618320610687e-05,
+      "loss": 0.5501,
+      "step": 13200
+    },
+    {
+      "epoch": 23.58,
+      "learning_rate": 4.267557251908397e-05,
+      "loss": 0.5429,
+      "step": 13300
+    },
+    {
+      "epoch": 23.76,
+      "learning_rate": 4.238931297709923e-05,
+      "loss": 0.5455,
+      "step": 13400
+    },
+    {
+      "epoch": 23.94,
+      "learning_rate": 4.21030534351145e-05,
+      "loss": 0.5504,
+      "step": 13500
+    },
+    {
+      "epoch": 23.94,
+      "eval_loss": 0.10192937403917313,
+      "eval_runtime": 31.0113,
+      "eval_samples_per_second": 5.74,
+      "eval_steps_per_second": 1.935,
+      "eval_wer": 0.11051394020826336,
+      "step": 13500
+    },
+    {
+      "epoch": 24.11,
+      "learning_rate": 4.181679389312977e-05,
+      "loss": 0.5601,
+      "step": 13600
+    },
+    {
+      "epoch": 24.29,
+      "learning_rate": 4.1530534351145035e-05,
+      "loss": 0.5419,
+      "step": 13700
+    },
+    {
+      "epoch": 24.47,
+      "learning_rate": 4.1244274809160305e-05,
+      "loss": 0.5389,
+      "step": 13800
+    },
+    {
+      "epoch": 24.65,
+      "learning_rate": 4.095801526717557e-05,
+      "loss": 0.5572,
+      "step": 13900
+    },
+    {
+      "epoch": 24.82,
+      "learning_rate": 4.067175572519084e-05,
+      "loss": 0.5476,
+      "step": 14000
+    },
+    {
+      "epoch": 24.82,
+      "eval_loss": 0.10260963439941406,
+      "eval_runtime": 31.1565,
+      "eval_samples_per_second": 5.713,
+      "eval_steps_per_second": 1.926,
+      "eval_wer": 0.11656029559959691,
+      "step": 14000
+    },
+    {
+      "epoch": 25.0,
+      "learning_rate": 4.03854961832061e-05,
+      "loss": 0.5554,
+      "step": 14100
+    },
+    {
+      "epoch": 25.18,
+      "learning_rate": 4.009923664122137e-05,
+      "loss": 0.537,
+      "step": 14200
+    },
+    {
+      "epoch": 25.35,
+      "learning_rate": 3.9812977099236635e-05,
+      "loss": 0.5346,
+      "step": 14300
+    },
+    {
+      "epoch": 25.53,
+      "learning_rate": 3.9526717557251906e-05,
+      "loss": 0.5245,
+      "step": 14400
+    },
+    {
+      "epoch": 25.71,
+      "learning_rate": 3.9240458015267176e-05,
+      "loss": 0.5375,
+      "step": 14500
+    },
+    {
+      "epoch": 25.71,
+      "eval_loss": 0.1107296496629715,
+      "eval_runtime": 30.9057,
+      "eval_samples_per_second": 5.759,
+      "eval_steps_per_second": 1.941,
+      "eval_wer": 0.11891165602955996,
+      "step": 14500
+    },
+    {
+      "epoch": 25.89,
+      "learning_rate": 3.895419847328244e-05,
+      "loss": 0.5437,
+      "step": 14600
+    },
+    {
+      "epoch": 26.06,
+      "learning_rate": 3.866793893129771e-05,
+      "loss": 0.5284,
+      "step": 14700
+    },
+    {
+      "epoch": 26.24,
+      "learning_rate": 3.838167938931297e-05,
+      "loss": 0.5276,
+      "step": 14800
+    },
+    {
+      "epoch": 26.42,
+      "learning_rate": 3.809541984732824e-05,
+      "loss": 0.5441,
+      "step": 14900
+    },
+    {
+      "epoch": 26.6,
+      "learning_rate": 3.7809160305343507e-05,
+      "loss": 0.5318,
+      "step": 15000
+    },
+    {
+      "epoch": 26.6,
+      "eval_loss": 0.10514429956674576,
+      "eval_runtime": 31.3716,
+      "eval_samples_per_second": 5.674,
+      "eval_steps_per_second": 1.913,
+      "eval_wer": 0.11420893516963386,
+      "step": 15000
+    },
+    {
+      "epoch": 26.77,
+      "learning_rate": 3.752290076335878e-05,
+      "loss": 0.5343,
+      "step": 15100
+    },
+    {
+      "epoch": 26.95,
+      "learning_rate": 3.723664122137404e-05,
+      "loss": 0.5394,
+      "step": 15200
+    },
+    {
+      "epoch": 27.13,
+      "learning_rate": 3.695038167938931e-05,
+      "loss": 0.5352,
+      "step": 15300
+    },
+    {
+      "epoch": 27.3,
+      "learning_rate": 3.6664122137404574e-05,
+      "loss": 0.5398,
+      "step": 15400
+    },
+    {
+      "epoch": 27.48,
+      "learning_rate": 3.6377862595419844e-05,
+      "loss": 0.5278,
+      "step": 15500
+    },
+    {
+      "epoch": 27.48,
+      "eval_loss": 0.10493182390928268,
+      "eval_runtime": 31.3849,
+      "eval_samples_per_second": 5.672,
+      "eval_steps_per_second": 1.912,
+      "eval_wer": 0.11656029559959691,
+      "step": 15500
+    },
+    {
+      "epoch": 27.66,
+      "learning_rate": 3.609160305343511e-05,
+      "loss": 0.538,
+      "step": 15600
+    },
+    {
+      "epoch": 27.84,
+      "learning_rate": 3.580534351145038e-05,
+      "loss": 0.5143,
+      "step": 15700
+    },
+    {
+      "epoch": 28.01,
+      "learning_rate": 3.551908396946565e-05,
+      "loss": 0.5229,
+      "step": 15800
+    },
+    {
+      "epoch": 28.19,
+      "learning_rate": 3.523282442748091e-05,
+      "loss": 0.522,
+      "step": 15900
+    },
+    {
+      "epoch": 28.37,
+      "learning_rate": 3.494656488549618e-05,
+      "loss": 0.5204,
+      "step": 16000
+    },
+    {
+      "epoch": 28.37,
+      "eval_loss": 0.10806475579738617,
+      "eval_runtime": 31.4114,
+      "eval_samples_per_second": 5.667,
+      "eval_steps_per_second": 1.91,
+      "eval_wer": 0.11823983876385623,
+      "step": 16000
+    },
+    {
+      "epoch": 28.55,
+      "learning_rate": 3.4660305343511445e-05,
+      "loss": 0.5322,
+      "step": 16100
+    },
+    {
+      "epoch": 28.72,
+      "learning_rate": 3.4374045801526715e-05,
+      "loss": 0.5151,
+      "step": 16200
+    },
+    {
+      "epoch": 28.9,
+      "learning_rate": 3.408778625954198e-05,
+      "loss": 0.5393,
+      "step": 16300
+    },
+    {
+      "epoch": 29.08,
+      "learning_rate": 3.380152671755725e-05,
+      "loss": 0.513,
+      "step": 16400
+    },
+    {
+      "epoch": 29.26,
+      "learning_rate": 3.351526717557251e-05,
+      "loss": 0.512,
+      "step": 16500
+    },
+    {
+      "epoch": 29.26,
+      "eval_loss": 0.10623880475759506,
+      "eval_runtime": 31.0174,
+      "eval_samples_per_second": 5.739,
+      "eval_steps_per_second": 1.934,
+      "eval_wer": 0.11555256970104132,
+      "step": 16500
+    },
+    {
+      "epoch": 29.43,
+      "learning_rate": 3.322900763358778e-05,
+      "loss": 0.5258,
+      "step": 16600
+    },
+    {
+      "epoch": 29.61,
+      "learning_rate": 3.294274809160305e-05,
+      "loss": 0.5366,
+      "step": 16700
+    },
+    {
+      "epoch": 29.79,
+      "learning_rate": 3.2656488549618316e-05,
+      "loss": 0.515,
+      "step": 16800
+    },
+    {
+      "epoch": 29.96,
+      "learning_rate": 3.2370229007633586e-05,
+      "loss": 0.5246,
+      "step": 16900
+    },
+    {
+      "epoch": 30.14,
+      "learning_rate": 3.208396946564885e-05,
+      "loss": 0.5082,
+      "step": 17000
+    },
+    {
+      "epoch": 30.14,
+      "eval_loss": 0.10452543944120407,
+      "eval_runtime": 30.9393,
+      "eval_samples_per_second": 5.753,
+      "eval_steps_per_second": 1.939,
+      "eval_wer": 0.11353711790393013,
+      "step": 17000
+    },
+    {
+      "epoch": 30.32,
+      "learning_rate": 3.179770992366412e-05,
+      "loss": 0.5029,
+      "step": 17100
+    },
+    {
+      "epoch": 30.5,
+      "learning_rate": 3.151431297709923e-05,
+      "loss": 0.5219,
+      "step": 17200
+    },
+    {
+      "epoch": 30.67,
+      "learning_rate": 3.12280534351145e-05,
+      "loss": 0.5235,
+      "step": 17300
+    },
+    {
+      "epoch": 30.85,
+      "learning_rate": 3.0941793893129764e-05,
+      "loss": 0.5141,
+      "step": 17400
+    },
+    {
+      "epoch": 31.03,
+      "learning_rate": 3.0655534351145035e-05,
+      "loss": 0.5193,
+      "step": 17500
+    },
+    {
+      "epoch": 31.03,
+      "eval_loss": 0.10911107808351517,
+      "eval_runtime": 30.9608,
+      "eval_samples_per_second": 5.749,
+      "eval_steps_per_second": 1.938,
+      "eval_wer": 0.11454484380248572,
+      "step": 17500
+    },
+    {
+      "epoch": 31.21,
+      "learning_rate": 3.0369274809160305e-05,
+      "loss": 0.5038,
+      "step": 17600
+    },
+    {
+      "epoch": 31.38,
+      "learning_rate": 3.008301526717557e-05,
+      "loss": 0.5086,
+      "step": 17700
+    },
+    {
+      "epoch": 31.56,
+      "learning_rate": 2.979675572519084e-05,
+      "loss": 0.5001,
+      "step": 17800
+    },
+    {
+      "epoch": 31.74,
+      "learning_rate": 2.9510496183206105e-05,
+      "loss": 0.5206,
+      "step": 17900
+    },
+    {
+      "epoch": 31.91,
+      "learning_rate": 2.9224236641221372e-05,
+      "loss": 0.5129,
+      "step": 18000
+    },
+    {
+      "epoch": 31.91,
+      "eval_loss": 0.10398419201374054,
+      "eval_runtime": 30.9924,
+      "eval_samples_per_second": 5.743,
+      "eval_steps_per_second": 1.936,
+      "eval_wer": 0.10883439704400404,
+      "step": 18000
+    },
+    {
+      "epoch": 32.09,
+      "learning_rate": 2.893797709923664e-05,
+      "loss": 0.5105,
+      "step": 18100
+    },
+    {
+      "epoch": 32.27,
+      "learning_rate": 2.8651717557251906e-05,
+      "loss": 0.5062,
+      "step": 18200
+    },
+    {
+      "epoch": 32.45,
+      "learning_rate": 2.8365458015267172e-05,
+      "loss": 0.5021,
+      "step": 18300
+    },
+    {
+      "epoch": 32.62,
+      "learning_rate": 2.807919847328244e-05,
+      "loss": 0.5122,
+      "step": 18400
+    },
+    {
+      "epoch": 32.8,
+      "learning_rate": 2.779293893129771e-05,
+      "loss": 0.5126,
+      "step": 18500
+    },
+    {
+      "epoch": 32.8,
+      "eval_loss": 0.10847991704940796,
+      "eval_runtime": 30.8721,
+      "eval_samples_per_second": 5.766,
+      "eval_steps_per_second": 1.944,
+      "eval_wer": 0.11689620423244877,
+      "step": 18500
+    },
+    {
+      "epoch": 32.98,
+      "learning_rate": 2.7506679389312976e-05,
+      "loss": 0.5179,
+      "step": 18600
+    },
+    {
+      "epoch": 33.16,
+      "learning_rate": 2.7220419847328243e-05,
+      "loss": 0.5049,
+      "step": 18700
+    },
+    {
+      "epoch": 33.33,
+      "learning_rate": 2.693416030534351e-05,
+      "loss": 0.5173,
+      "step": 18800
+    },
+    {
+      "epoch": 33.51,
+      "learning_rate": 2.6647900763358777e-05,
+      "loss": 0.5115,
+      "step": 18900
+    },
+    {
+      "epoch": 33.69,
+      "learning_rate": 2.6361641221374043e-05,
+      "loss": 0.496,
+      "step": 19000
+    },
+    {
+      "epoch": 33.69,
+      "eval_loss": 0.10698471963405609,
+      "eval_runtime": 31.005,
+      "eval_samples_per_second": 5.741,
+      "eval_steps_per_second": 1.935,
+      "eval_wer": 0.11656029559959691,
+      "step": 19000
+    },
+    {
+      "epoch": 33.86,
+      "learning_rate": 2.607538167938931e-05,
+      "loss": 0.4938,
+      "step": 19100
+    },
+    {
+      "epoch": 34.04,
+      "learning_rate": 2.5789122137404577e-05,
+      "loss": 0.5128,
+      "step": 19200
+    },
+    {
+      "epoch": 34.22,
+      "learning_rate": 2.550572519083969e-05,
+      "loss": 0.497,
+      "step": 19300
+    },
+    {
+      "epoch": 34.4,
+      "learning_rate": 2.521946564885496e-05,
+      "loss": 0.4879,
+      "step": 19400
+    },
+    {
+      "epoch": 34.57,
+      "learning_rate": 2.493320610687023e-05,
+      "loss": 0.5017,
+      "step": 19500
+    },
+    {
+      "epoch": 34.57,
+      "eval_loss": 0.11190272867679596,
+      "eval_runtime": 30.8238,
+      "eval_samples_per_second": 5.775,
+      "eval_steps_per_second": 1.947,
+      "eval_wer": 0.11622438696674504,
+      "step": 19500
+    },
+    {
+      "epoch": 34.75,
+      "learning_rate": 2.4646946564885495e-05,
+      "loss": 0.4994,
+      "step": 19600
+    },
+    {
+      "epoch": 34.93,
+      "learning_rate": 2.436354961832061e-05,
+      "loss": 0.4973,
+      "step": 19700
+    },
+    {
+      "epoch": 35.11,
+      "learning_rate": 2.4077290076335876e-05,
+      "loss": 0.5012,
+      "step": 19800
+    },
+    {
+      "epoch": 35.28,
+      "learning_rate": 2.3791030534351143e-05,
+      "loss": 0.4889,
+      "step": 19900
+    },
+    {
+      "epoch": 35.46,
+      "learning_rate": 2.350477099236641e-05,
+      "loss": 0.4808,
+      "step": 20000
+    },
+    {
+      "epoch": 35.46,
+      "eval_loss": 0.11005562543869019,
+      "eval_runtime": 31.0411,
+      "eval_samples_per_second": 5.734,
+      "eval_steps_per_second": 1.933,
+      "eval_wer": 0.113873026536782,
+      "step": 20000
+    },
+    {
+      "epoch": 35.64,
+      "learning_rate": 2.3218511450381677e-05,
+      "loss": 0.4984,
+      "step": 20100
+    },
+    {
+      "epoch": 35.82,
+      "learning_rate": 2.2932251908396944e-05,
+      "loss": 0.4929,
+      "step": 20200
+    },
+    {
+      "epoch": 35.99,
+      "learning_rate": 2.2645992366412214e-05,
+      "loss": 0.4861,
+      "step": 20300
+    },
+    {
+      "epoch": 36.17,
+      "learning_rate": 2.235973282442748e-05,
+      "loss": 0.506,
+      "step": 20400
+    },
+    {
+      "epoch": 36.35,
+      "learning_rate": 2.2073473282442747e-05,
+      "loss": 0.4939,
+      "step": 20500
+    },
+    {
+      "epoch": 36.35,
+      "eval_loss": 0.10813739150762558,
+      "eval_runtime": 31.0724,
+      "eval_samples_per_second": 5.729,
+      "eval_steps_per_second": 1.931,
+      "eval_wer": 0.11252939200537454,
+      "step": 20500
+    },
+    {
+      "epoch": 36.52,
+      "learning_rate": 2.1787213740458014e-05,
+      "loss": 0.4789,
+      "step": 20600
+    },
+    {
+      "epoch": 36.7,
+      "learning_rate": 2.150095419847328e-05,
+      "loss": 0.4904,
+      "step": 20700
+    },
+    {
+      "epoch": 36.88,
+      "learning_rate": 2.1214694656488548e-05,
+      "loss": 0.4916,
+      "step": 20800
+    },
+    {
+      "epoch": 37.06,
+      "learning_rate": 2.0928435114503815e-05,
+      "loss": 0.487,
+      "step": 20900
+    },
+    {
+      "epoch": 37.23,
+      "learning_rate": 2.064217557251908e-05,
+      "loss": 0.4738,
+      "step": 21000
+    },
+    {
+      "epoch": 37.23,
+      "eval_loss": 0.10911141335964203,
+      "eval_runtime": 31.0688,
+      "eval_samples_per_second": 5.729,
+      "eval_steps_per_second": 1.931,
+      "eval_wer": 0.10984212294255963,
+      "step": 21000
+    },
+    {
+      "epoch": 37.41,
+      "learning_rate": 2.0355916030534352e-05,
+      "loss": 0.4757,
+      "step": 21100
+    },
+    {
+      "epoch": 37.59,
+      "learning_rate": 2.006965648854962e-05,
+      "loss": 0.4752,
+      "step": 21200
+    },
+    {
+      "epoch": 37.77,
+      "learning_rate": 1.9783396946564885e-05,
+      "loss": 0.4939,
+      "step": 21300
+    },
+    {
+      "epoch": 37.94,
+      "learning_rate": 1.9497137404580152e-05,
+      "loss": 0.4673,
+      "step": 21400
+    },
+    {
+      "epoch": 38.12,
+      "learning_rate": 1.921087786259542e-05,
+      "loss": 0.4978,
+      "step": 21500
+    },
+    {
+      "epoch": 38.12,
+      "eval_loss": 0.10570317506790161,
+      "eval_runtime": 31.0697,
+      "eval_samples_per_second": 5.729,
+      "eval_steps_per_second": 1.931,
+      "eval_wer": 0.1091703056768559,
+      "step": 21500
+    },
+    {
+      "epoch": 38.3,
+      "learning_rate": 1.8924618320610686e-05,
+      "loss": 0.4829,
+      "step": 21600
+    },
+    {
+      "epoch": 38.47,
+      "learning_rate": 1.8638358778625953e-05,
+      "loss": 0.4857,
+      "step": 21700
+    },
+    {
+      "epoch": 38.65,
+      "learning_rate": 1.835209923664122e-05,
+      "loss": 0.4618,
+      "step": 21800
+    },
+    {
+      "epoch": 38.83,
+      "learning_rate": 1.8065839694656486e-05,
+      "loss": 0.4831,
+      "step": 21900
+    },
+    {
+      "epoch": 39.01,
+      "learning_rate": 1.77824427480916e-05,
+      "loss": 0.4972,
+      "step": 22000
+    },
+    {
+      "epoch": 39.01,
+      "eval_loss": 0.10742757469415665,
+      "eval_runtime": 31.1659,
+      "eval_samples_per_second": 5.711,
+      "eval_steps_per_second": 1.925,
+      "eval_wer": 0.11051394020826336,
+      "step": 22000
+    },
+    {
+      "epoch": 39.18,
+      "learning_rate": 1.7496183206106867e-05,
+      "loss": 0.4974,
+      "step": 22100
+    },
+    {
+      "epoch": 39.36,
+      "learning_rate": 1.7209923664122138e-05,
+      "loss": 0.4778,
+      "step": 22200
+    },
+    {
+      "epoch": 39.54,
+      "learning_rate": 1.6923664122137404e-05,
+      "loss": 0.4863,
+      "step": 22300
+    },
+    {
+      "epoch": 39.72,
+      "learning_rate": 1.663740458015267e-05,
+      "loss": 0.4815,
+      "step": 22400
+    },
+    {
+      "epoch": 39.89,
+      "learning_rate": 1.6351145038167938e-05,
+      "loss": 0.4773,
+      "step": 22500
+    },
+    {
+      "epoch": 39.89,
+      "eval_loss": 0.10622948408126831,
+      "eval_runtime": 31.5999,
+      "eval_samples_per_second": 5.633,
+      "eval_steps_per_second": 1.899,
+      "eval_wer": 0.11084984884111522,
+      "step": 22500
+    },
+    {
+      "epoch": 40.07,
+      "learning_rate": 1.6064885496183205e-05,
+      "loss": 0.4808,
+      "step": 22600
+    },
+    {
+      "epoch": 40.25,
+      "learning_rate": 1.577862595419847e-05,
+      "loss": 0.4804,
+      "step": 22700
+    },
+    {
+      "epoch": 40.43,
+      "learning_rate": 1.549236641221374e-05,
+      "loss": 0.4796,
+      "step": 22800
+    },
+    {
+      "epoch": 40.6,
+      "learning_rate": 1.5206106870229005e-05,
+      "loss": 0.4866,
+      "step": 22900
+    },
+    {
+      "epoch": 40.78,
+      "learning_rate": 1.4919847328244272e-05,
+      "loss": 0.4741,
+      "step": 23000
+    },
+    {
+      "epoch": 40.78,
+      "eval_loss": 0.10567349940538406,
+      "eval_runtime": 31.2551,
+      "eval_samples_per_second": 5.695,
+      "eval_steps_per_second": 1.92,
+      "eval_wer": 0.10849848841115217,
+      "step": 23000
+    },
+    {
+      "epoch": 40.96,
+      "learning_rate": 1.4633587786259542e-05,
+      "loss": 0.4976,
+      "step": 23100
+    },
+    {
+      "epoch": 41.13,
+      "learning_rate": 1.4347328244274809e-05,
+      "loss": 0.4811,
+      "step": 23200
+    },
+    {
+      "epoch": 41.31,
+      "learning_rate": 1.4061068702290076e-05,
+      "loss": 0.4904,
+      "step": 23300
+    },
+    {
+      "epoch": 41.49,
+      "learning_rate": 1.3774809160305341e-05,
+      "loss": 0.4658,
+      "step": 23400
+    },
+    {
+      "epoch": 41.67,
+      "learning_rate": 1.3488549618320608e-05,
+      "loss": 0.4776,
+      "step": 23500
+    },
+    {
+      "epoch": 41.67,
+      "eval_loss": 0.10771991312503815,
+      "eval_runtime": 31.4029,
+      "eval_samples_per_second": 5.668,
+      "eval_steps_per_second": 1.911,
+      "eval_wer": 0.10849848841115217,
+      "step": 23500
+    },
+    {
+      "epoch": 41.84,
+      "learning_rate": 1.3202290076335878e-05,
+      "loss": 0.4745,
+      "step": 23600
+    },
+    {
+      "epoch": 42.02,
+      "learning_rate": 1.2916030534351145e-05,
+      "loss": 0.4746,
+      "step": 23700
+    },
+    {
+      "epoch": 42.2,
+      "learning_rate": 1.262977099236641e-05,
+      "loss": 0.4858,
+      "step": 23800
+    },
+    {
+      "epoch": 42.38,
+      "learning_rate": 1.2343511450381677e-05,
+      "loss": 0.4687,
+      "step": 23900
+    },
+    {
+      "epoch": 42.55,
+      "learning_rate": 1.2057251908396947e-05,
+      "loss": 0.4637,
+      "step": 24000
+    },
+    {
+      "epoch": 42.55,
+      "eval_loss": 0.10607700049877167,
+      "eval_runtime": 31.045,
+      "eval_samples_per_second": 5.734,
+      "eval_steps_per_second": 1.933,
+      "eval_wer": 0.10950621430970776,
+      "step": 24000
+    },
+    {
+      "epoch": 42.73,
+      "learning_rate": 1.1770992366412214e-05,
+      "loss": 0.4462,
+      "step": 24100
+    },
+    {
+      "epoch": 42.91,
+      "learning_rate": 1.1484732824427479e-05,
+      "loss": 0.4815,
+      "step": 24200
+    },
+    {
+      "epoch": 43.09,
+      "learning_rate": 1.1198473282442746e-05,
+      "loss": 0.461,
+      "step": 24300
+    },
+    {
+      "epoch": 43.26,
+      "learning_rate": 1.0912213740458016e-05,
+      "loss": 0.4594,
+      "step": 24400
+    },
+    {
+      "epoch": 43.44,
+      "learning_rate": 1.0625954198473283e-05,
+      "loss": 0.4853,
+      "step": 24500
+    },
+    {
+      "epoch": 43.44,
+      "eval_loss": 0.10806521028280258,
+      "eval_runtime": 31.4451,
+      "eval_samples_per_second": 5.661,
+      "eval_steps_per_second": 1.908,
+      "eval_wer": 0.10749076251259658,
+      "step": 24500
+    },
+    {
+      "epoch": 43.62,
+      "learning_rate": 1.033969465648855e-05,
+      "loss": 0.4747,
+      "step": 24600
+    },
+    {
+      "epoch": 43.79,
+      "learning_rate": 1.0053435114503815e-05,
+      "loss": 0.4649,
+      "step": 24700
+    },
+    {
+      "epoch": 43.97,
+      "learning_rate": 9.767175572519081e-06,
+      "loss": 0.471,
+      "step": 24800
+    },
+    {
+      "epoch": 44.15,
+      "learning_rate": 9.480916030534352e-06,
+      "loss": 0.453,
+      "step": 24900
+    },
+    {
+      "epoch": 44.33,
+      "learning_rate": 9.194656488549618e-06,
+      "loss": 0.4602,
+      "step": 25000
+    },
+    {
+      "epoch": 44.33,
+      "eval_loss": 0.1075613871216774,
+      "eval_runtime": 31.217,
+      "eval_samples_per_second": 5.702,
+      "eval_steps_per_second": 1.922,
+      "eval_wer": 0.10849848841115217,
+      "step": 25000
+    },
+    {
+      "epoch": 44.5,
+      "learning_rate": 8.908396946564884e-06,
+      "loss": 0.4579,
+      "step": 25100
+    },
+    {
+      "epoch": 44.68,
+      "learning_rate": 8.622137404580152e-06,
+      "loss": 0.4752,
+      "step": 25200
+    },
+    {
+      "epoch": 44.86,
+      "learning_rate": 8.335877862595419e-06,
+      "loss": 0.4708,
+      "step": 25300
+    },
+    {
+      "epoch": 45.04,
+      "learning_rate": 8.049618320610687e-06,
+      "loss": 0.4613,
+      "step": 25400
+    },
+    {
+      "epoch": 45.21,
+      "learning_rate": 7.763358778625954e-06,
+      "loss": 0.4667,
+      "step": 25500
+    },
+    {
+      "epoch": 45.21,
+      "eval_loss": 0.1077902689576149,
+      "eval_runtime": 31.0572,
+      "eval_samples_per_second": 5.731,
+      "eval_steps_per_second": 1.932,
+      "eval_wer": 0.10782667114544844,
+      "step": 25500
+    },
+    {
+      "epoch": 45.39,
+      "learning_rate": 7.477099236641221e-06,
+      "loss": 0.4709,
+      "step": 25600
+    },
+    {
+      "epoch": 45.57,
+      "learning_rate": 7.190839694656488e-06,
+      "loss": 0.4523,
+      "step": 25700
+    },
+    {
+      "epoch": 45.74,
+      "learning_rate": 6.9045801526717555e-06,
+      "loss": 0.4537,
+      "step": 25800
+    },
+    {
+      "epoch": 45.92,
+      "learning_rate": 6.62118320610687e-06,
+      "loss": 0.4642,
+      "step": 25900
+    },
+    {
+      "epoch": 46.1,
+      "learning_rate": 6.334923664122137e-06,
+      "loss": 0.4484,
+      "step": 26000
+    },
+    {
+      "epoch": 46.1,
+      "eval_loss": 0.10564640909433365,
+      "eval_runtime": 31.5659,
+      "eval_samples_per_second": 5.639,
+      "eval_steps_per_second": 1.901,
+      "eval_wer": 0.10816257977830031,
+      "step": 26000
+    },
+    {
+      "epoch": 46.28,
+      "learning_rate": 6.048664122137404e-06,
+      "loss": 0.4505,
+      "step": 26100
+    },
+    {
+      "epoch": 46.45,
+      "learning_rate": 5.762404580152671e-06,
+      "loss": 0.4758,
+      "step": 26200
+    },
+    {
+      "epoch": 46.63,
+      "learning_rate": 5.476145038167939e-06,
+      "loss": 0.4528,
+      "step": 26300
+    },
+    {
+      "epoch": 46.81,
+      "learning_rate": 5.1898854961832056e-06,
+      "loss": 0.4579,
+      "step": 26400
+    },
+    {
+      "epoch": 46.99,
+      "learning_rate": 4.903625954198473e-06,
+      "loss": 0.4601,
+      "step": 26500
+    },
+    {
+      "epoch": 46.99,
+      "eval_loss": 0.10663167387247086,
+      "eval_runtime": 31.3245,
+      "eval_samples_per_second": 5.682,
+      "eval_steps_per_second": 1.915,
+      "eval_wer": 0.10782667114544844,
+      "step": 26500
+    },
+    {
+      "epoch": 47.16,
+      "learning_rate": 4.61736641221374e-06,
+      "loss": 0.4634,
+      "step": 26600
+    },
+    {
+      "epoch": 47.34,
+      "learning_rate": 4.331106870229007e-06,
+      "loss": 0.4603,
+      "step": 26700
+    },
+    {
+      "epoch": 47.52,
+      "learning_rate": 4.0448473282442745e-06,
+      "loss": 0.4653,
+      "step": 26800
+    },
+    {
+      "epoch": 47.69,
+      "learning_rate": 3.7585877862595418e-06,
+      "loss": 0.4619,
+      "step": 26900
+    },
+    {
+      "epoch": 47.87,
+      "learning_rate": 3.472328244274809e-06,
+      "loss": 0.4691,
+      "step": 27000
+    },
+    {
+      "epoch": 47.87,
+      "eval_loss": 0.10679604858160019,
+      "eval_runtime": 31.1673,
+      "eval_samples_per_second": 5.711,
+      "eval_steps_per_second": 1.925,
+      "eval_wer": 0.10849848841115217,
+      "step": 27000
+    },
+    {
+      "epoch": 48.05,
+      "learning_rate": 3.1860687022900762e-06,
+      "loss": 0.4582,
+      "step": 27100
+    },
+    {
+      "epoch": 48.23,
+      "learning_rate": 2.8998091603053435e-06,
+      "loss": 0.4439,
+      "step": 27200
+    },
+    {
+      "epoch": 48.4,
+      "learning_rate": 2.6135496183206107e-06,
+      "loss": 0.4617,
+      "step": 27300
+    },
+    {
+      "epoch": 48.58,
+      "learning_rate": 2.3272900763358775e-06,
+      "loss": 0.4538,
+      "step": 27400
+    },
+    {
+      "epoch": 48.76,
+      "learning_rate": 2.0410305343511447e-06,
+      "loss": 0.4457,
+      "step": 27500
+    },
+    {
+      "epoch": 48.76,
+      "eval_loss": 0.10663049668073654,
+      "eval_runtime": 31.2143,
+      "eval_samples_per_second": 5.703,
+      "eval_steps_per_second": 1.922,
+      "eval_wer": 0.10782667114544844,
+      "step": 27500
+    },
+    {
+      "epoch": 48.94,
+      "learning_rate": 1.7547709923664122e-06,
+      "loss": 0.4766,
+      "step": 27600
+    },
+    {
+      "epoch": 49.11,
+      "learning_rate": 1.4685114503816792e-06,
+      "loss": 0.4576,
+      "step": 27700
+    },
+    {
+      "epoch": 49.29,
+      "learning_rate": 1.185114503816794e-06,
+      "loss": 0.4616,
+      "step": 27800
+    },
+    {
+      "epoch": 49.47,
+      "learning_rate": 8.98854961832061e-07,
+      "loss": 0.4588,
+      "step": 27900
+    },
+    {
+      "epoch": 49.65,
+      "learning_rate": 6.125954198473282e-07,
+      "loss": 0.475,
+      "step": 28000
+    },
+    {
+      "epoch": 49.65,
+      "eval_loss": 0.1059938296675682,
+      "eval_runtime": 32.0006,
+      "eval_samples_per_second": 5.562,
+      "eval_steps_per_second": 1.875,
+      "eval_wer": 0.10816257977830031,
+      "step": 28000
+    },
+    {
+      "epoch": 49.82,
+      "learning_rate": 3.263358778625954e-07,
+      "loss": 0.4622,
+      "step": 28100
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 4.0076335877862596e-08,
+      "loss": 0.4567,
+      "step": 28200
+    },
+    {
+      "epoch": 50.0,
+      "step": 28200,
+      "total_flos": 6.992186583697227e+19,
+      "train_loss": 0.6889870901987062,
+      "train_runtime": 132713.9902,
+      "train_samples_per_second": 2.551,
+      "train_steps_per_second": 0.212
+    }
+  ],
+  "max_steps": 28200,
+  "num_train_epochs": 50,
+  "total_flos": 6.992186583697227e+19,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95b1c12bb3e43e845c3fafc74da0962a2f160b7c3a7e2ee8805288861ea2bbfe
+size 3183

vocab.json

@@ -0,0 +1 @@
+{"&": 1, "'": 2, "(": 3, ")": 4, "/": 5, "0": 6, "1": 7, "2": 8, "3": 9, "4": 10, "5": 11, "7": 12, "9": 13, "a": 14, "b": 15, "c": 16, "d": 17, "e": 18, "f": 19, "g": 20, "h": 21, "i": 22, "j": 23, "k": 24, "l": 25, "m": 26, "n": 27, "o": 28, "p": 29, "q": 30, "r": 31, "s": 32, "t": 33, "u": 34, "v": 35, "w": 36, "x": 37, "y": 38, "z": 39, "ß": 40, "à": 41, "â": 42, "ä": 43, "ç": 44, "è": 45, "é": 46, "ê": 47, "ë": 48, "î": 49, "ï": 50, "ô": 51, "ö": 52, "û": 53, "ü": 54, "|": 0, "[UNK]": 55, "[PAD]": 56}