Uploaded and Created

README.md

@@ -0,0 +1,171 @@
+---
+language: en
+license: apache-2.0
+tags:
+- phoneme-recognition
+- generated_from_trainer
+datasets:
+- w11wo/ljspeech_phonemes
+base_model: Wav2Vec2-Base
+inference:
+  parameters:
+    function_to_apply: none
+model-index:
+- name: Wav2Vec2 LJSpeech Gruut
+  results:
+  - task:
+      type: automatic-speech-recognition
+      name: Automatic Speech Recognition
+    dataset:
+      name: LJSpeech
+      type: ljspeech_phonemes
+    metrics:
+    - type: per
+      value: 0.0099
+      name: Test PER (w/o stress)
+    - type: cer
+      value: 0.0058
+      name: Test CER (w/o stress)
+---
+
+# Wav2Vec2 LJSpeech Gruut
+
+Clone of wav2vec2-ljspeech-gruut because I want to use pipeline and get the logits from it
+
+## Model
+
+| Model                     | #params | Arch.       | Training/Validation data (text) |
+| ------------------------- | ------- | ----------- | ------------------------------- |
+| `wav2vec2-ljspeech-gruut` | 94M     | wav2vec 2.0 | `LJSpech Phonemes` Dataset      |
+
+## Evaluation Results
+
+The model achieves the following results on evaluation:
+
+| Dataset                      | PER (w/o stress) | CER (w/o stress) |
+| ---------------------------- | :--------------: | :--------------: |
+| `LJSpech Phonemes` Test Data |      0.99%       |      0.58%       |
+
+## Usage
+
+```py
+from transformers import AutoProcessor, AutoModelForCTC, Wav2Vec2Processor
+import librosa
+import torch
+from itertools import groupby
+from datasets import load_dataset
+
+def decode_phonemes(
+    ids: torch.Tensor, processor: Wav2Vec2Processor, ignore_stress: bool = False
+) -> str:
+    """CTC-like decoding. First removes consecutive duplicates, then removes special tokens."""
+    # removes consecutive duplicates
+    ids = [id_ for id_, _ in groupby(ids)]
+
+    special_token_ids = processor.tokenizer.all_special_ids + [
+        processor.tokenizer.word_delimiter_token_id
+    ]
+    # converts id to token, skipping special tokens
+    phonemes = [processor.decode(id_) for id_ in ids if id_ not in special_token_ids]
+
+    # joins phonemes
+    prediction = " ".join(phonemes)
+
+    # whether to ignore IPA stress marks
+    if ignore_stress == True:
+        prediction = prediction.replace("ˈ", "").replace("ˌ", "")
+
+    return prediction
+
+checkpoint = "bookbot/wav2vec2-ljspeech-gruut"
+
+model = AutoModelForCTC.from_pretrained(checkpoint)
+processor = AutoProcessor.from_pretrained(checkpoint)
+sr = processor.feature_extractor.sampling_rate
+
+# load dummy dataset and read soundfiles
+ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
+audio_array = ds[0]["audio"]["array"]
+
+# or, read a single audio file
+# audio_array, _ = librosa.load("myaudio.wav", sr=sr)
+
+inputs = processor(audio_array, return_tensors="pt", padding=True)
+
+with torch.no_grad():
+    logits = model(inputs["input_values"]).logits
+
+predicted_ids = torch.argmax(logits, dim=-1)
+prediction = decode_phonemes(predicted_ids[0], processor, ignore_stress=True)
+# => should give 'b ɪ k ʌ z j u ɚ z s l i p ɪ ŋ ɪ n s t ɛ d ə v k ɔ ŋ k ɚ ɪ ŋ ð ə l ʌ v l i ɹ z p ɹ ɪ n s ə s h æ z b ɪ k ʌ m ə v f ɪ t ə l w ɪ θ n b oʊ p ɹ ə ʃ æ ɡ i s ɪ t s ð ɛ ɹ ə k u ɪ ŋ d ʌ v'
+```
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+
+- `learning_rate`: 0.0001
+- `train_batch_size`: 16
+- `eval_batch_size`: 8
+- `seed`: 42
+- `gradient_accumulation_steps`: 2
+- `total_train_batch_size`: 32
+- `optimizer`: Adam with `betas=(0.9,0.999)` and `epsilon=1e-08`
+- `lr_scheduler_type`: linear
+- `lr_scheduler_warmup_steps`: 1000
+- `num_epochs`: 30.0
+- `mixed_precision_training`: Native AMP
+
+### Training results
+
+| Training Loss | Epoch | Step  | Validation Loss |  Wer   |  Cer   |
+| :-----------: | :---: | :---: | :-------------: | :----: | :----: |
+|    No log     |  1.0  |  348  |     2.2818      |  1.0   |  1.0   |
+|    2.6692     |  2.0  |  696  |     0.2045      | 0.0527 | 0.0299 |
+|    0.2225     |  3.0  | 1044  |     0.1162      | 0.0319 | 0.0189 |
+|    0.2225     |  4.0  | 1392  |     0.0927      | 0.0235 | 0.0147 |
+|    0.0868     |  5.0  | 1740  |     0.0797      | 0.0218 | 0.0143 |
+|    0.0598     |  6.0  | 2088  |     0.0715      | 0.0197 | 0.0128 |
+|    0.0598     |  7.0  | 2436  |     0.0652      | 0.0160 | 0.0103 |
+|    0.0447     |  8.0  | 2784  |     0.0571      | 0.0152 | 0.0095 |
+|    0.0368     |  9.0  | 3132  |     0.0608      | 0.0163 | 0.0112 |
+|    0.0368     | 10.0  | 3480  |     0.0586      | 0.0137 | 0.0083 |
+|    0.0303     | 11.0  | 3828  |     0.0641      | 0.0141 | 0.0085 |
+|    0.0273     | 12.0  | 4176  |     0.0656      | 0.0131 | 0.0079 |
+|    0.0232     | 13.0  | 4524  |     0.0690      | 0.0133 | 0.0082 |
+|    0.0232     | 14.0  | 4872  |     0.0598      | 0.0128 | 0.0079 |
+|    0.0189     | 15.0  | 5220  |     0.0671      | 0.0121 | 0.0074 |
+|     0.017     | 16.0  | 5568  |     0.0654      | 0.0114 | 0.0069 |
+|     0.017     | 17.0  | 5916  |     0.0751      | 0.0118 | 0.0073 |
+|    0.0146     | 18.0  | 6264  |     0.0653      | 0.0112 | 0.0068 |
+|    0.0127     | 19.0  | 6612  |     0.0682      | 0.0112 | 0.0069 |
+|    0.0127     | 20.0  | 6960  |     0.0678      | 0.0114 | 0.0068 |
+|    0.0114     | 21.0  | 7308  |     0.0656      | 0.0111 | 0.0066 |
+|    0.0101     | 22.0  | 7656  |     0.0669      | 0.0109 | 0.0066 |
+|    0.0092     | 23.0  | 8004  |     0.0677      | 0.0108 | 0.0065 |
+|    0.0092     | 24.0  | 8352  |     0.0653      | 0.0104 | 0.0063 |
+|    0.0088     | 25.0  | 8700  |     0.0673      | 0.0102 | 0.0063 |
+|    0.0074     | 26.0  | 9048  |     0.0669      | 0.0105 | 0.0064 |
+|    0.0074     | 27.0  | 9396  |     0.0707      | 0.0101 | 0.0061 |
+|    0.0066     | 28.0  | 9744  |     0.0673      | 0.0100 | 0.0060 |
+|    0.0058     | 29.0  | 10092 |     0.0689      | 0.0100 | 0.0059 |
+|    0.0058     | 30.0  | 10440 |     0.0683      | 0.0099 | 0.0058 |
+
+
+## Disclaimer
+
+Do consider the biases which came from pre-training datasets that may be carried over into the results of this model.
+
+## Authors
+
+Wav2Vec2 LJSpeech Gruut was trained and evaluated by [Wilson Wongso](https://w11wo.github.io/). All computation and development are done on Google Cloud.
+
+## Framework versions
+
+- Transformers 4.26.0.dev0
+- Pytorch 1.10.0
+- Datasets 2.7.1
+- Tokenizers 0.13.2
+- Gruut 2.3.4

added_tokens.json

@@ -0,0 +1,4 @@
+{
+  "</s>": 44,
+  "<s>": 43
+}

all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 30.0,
+    "eval_cer": 0.0058143100249896515,
+    "eval_loss": 0.06832413375377655,
+    "eval_runtime": 87.0708,
+    "eval_samples": 1965,
+    "eval_samples_per_second": 22.568,
+    "eval_steps_per_second": 2.825,
+    "eval_wer": 0.009874807524938073,
+    "train_loss": 0.15939651108792915,
+    "train_runtime": 19958.548,
+    "train_samples": 11135,
+    "train_samples_per_second": 16.737,
+    "train_steps_per_second": 0.523
+}

config.json

@@ -0,0 +1,117 @@
+{
+  "_name_or_path": "facebook/wav2vec2-base",
+  "activation_dropout": 0.0,
+  "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": 256,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": false,
+  "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": false,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_norm": "group",
+  "feat_proj_dropout": 0.0,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "freeze_feat_extract_train": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.0,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.05,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "no_mask_channel_overlap": false,
+  "no_mask_time_overlap": false,
+  "num_adapter_layers": 3,
+  "num_attention_heads": 12,
+  "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": 12,
+  "num_negatives": 100,
+  "output_hidden_size": 768,
+  "pad_token_id": 42,
+  "proj_codevector_dim": 256,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.26.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 45,
+  "xvector_output_dim": 512
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36094944bb26c933f05f2d3e40e14ae495479700380597ce11e1546c6df08c93
+size 134

preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70b3f3f15a1a1f2ca80f9af5764e9b9098a0a7b7c5a6a43494edd2fcd9503dec
+size 134

run.sh

@@ -0,0 +1,29 @@
+python run_speech_recognition_ctc.py \
+    --dataset_name="w11wo/ljspeech_phonemes" \
+    --text_column_name="phonemes" \
+    --train_split_name="train" \
+    --model_name_or_path="facebook/wav2vec2-base" \
+    --output_dir="./wav2vec2-ljspeech-gruut" \
+    --overwrite_output_dir \
+    --num_train_epochs="30" \
+    --per_device_train_batch_size="16" \
+    --gradient_accumulation_steps="2" \
+    --learning_rate="1e-4" \
+    --warmup_steps="1000" \
+    --weight_decay="0.005" \
+    --evaluation_strategy="epoch" \
+    --eval_metrics wer cer \
+    --save_strategy="epoch" \
+    --layerdrop="0.0" \
+    --save_total_limit="3" \
+    --freeze_feature_encoder \
+    --gradient_checkpointing \
+    --chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � ˈ ˌ \
+    --fp16 \
+    --group_by_length \
+    --report_to="tensorboard" \
+    --push_to_hub \
+    --do_train --do_eval \
+    --hub_model_id="bookbot/wav2vec2-ljspeech-gruut" \
+    --hub_private_repo="True" \
+    --use_auth_token="True"

run_speech_recognition_ctc.py

@@ -0,0 +1,856 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+""" Fine-tuning a 🤗 Transformers CTC model for automatic speech recognition"""
+
+import functools
+import json
+import logging
+import os
+import re
+import sys
+import warnings
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional, Union
+
+import datasets
+import numpy as np
+import torch
+from datasets import DatasetDict, load_dataset
+
+import evaluate
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoFeatureExtractor,
+    AutoModelForCTC,
+    AutoProcessor,
+    AutoTokenizer,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    Wav2Vec2Processor,
+    set_seed,
+)
+from transformers.trainer_utils import get_last_checkpoint, is_main_process
+from transformers.utils import check_min_version, send_example_telemetry
+from transformers.utils.versions import require_version
+
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.26.0.dev0")
+
+require_version(
+    "datasets>=1.18.0",
+    "To fix: pip install -r examples/pytorch/speech-recognition/requirements.txt",
+)
+
+
+logger = logging.getLogger(__name__)
+
+
+def list_field(default=None, metadata=None):
+    return field(default_factory=lambda: default, metadata=metadata)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={
+            "help": "Path to pretrained model or model identifier from huggingface.co/models"
+        }
+    )
+    tokenizer_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"
+        },
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "Where do you want to store the pretrained models downloaded from huggingface.co"
+        },
+    )
+    freeze_feature_encoder: bool = field(
+        default=True,
+        metadata={"help": "Whether to freeze the feature encoder layers of the model."},
+    )
+    attention_dropout: float = field(
+        default=0.0,
+        metadata={"help": "The dropout ratio for the attention probabilities."},
+    )
+    activation_dropout: float = field(
+        default=0.0,
+        metadata={
+            "help": "The dropout ratio for activations inside the fully connected layer."
+        },
+    )
+    feat_proj_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for the projected features."}
+    )
+    hidden_dropout: float = field(
+        default=0.0,
+        metadata={
+            "help": "The dropout probability for all fully connected layers in the embeddings, encoder, and pooler."
+        },
+    )
+    final_dropout: float = field(
+        default=0.0,
+        metadata={"help": "The dropout probability for the final projection layer."},
+    )
+    mask_time_prob: float = field(
+        default=0.05,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the time axis to be chosen as the start of the vector"
+                "span to be masked. Approximately ``mask_time_prob * sequence_length // mask_time_length`` feature"
+                "vectors will be masked along the time axis."
+            )
+        },
+    )
+    mask_time_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the time axis."},
+    )
+    mask_feature_prob: float = field(
+        default=0.0,
+        metadata={
+            "help": (
+                "Probability of each feature vector along the feature axis to be chosen as the start of the vectorspan"
+                " to be masked. Approximately ``mask_feature_prob * sequence_length // mask_feature_length`` feature"
+                " bins will be masked along the time axis."
+            )
+        },
+    )
+    mask_feature_length: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the feature axis."},
+    )
+    layerdrop: float = field(
+        default=0.0, metadata={"help": "The LayerDrop probability."}
+    )
+    ctc_loss_reduction: Optional[str] = field(
+        default="mean",
+        metadata={
+            "help": "The way the ctc loss should be reduced. Should be one of 'mean' or 'sum'."
+        },
+    )
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    dataset_name: str = field(
+        metadata={
+            "help": "The configuration name of the dataset to use (via the datasets library)."
+        }
+    )
+    dataset_config_name: str = field(
+        default=None,
+        metadata={
+            "help": "The configuration name of the dataset to use (via the datasets library)."
+        },
+    )
+    train_split_name: str = field(
+        default="train+validation",
+        metadata={
+            "help": (
+                "The name of the training data set split to use (via the datasets library). Defaults to "
+                "'train+validation'"
+            )
+        },
+    )
+    eval_split_name: str = field(
+        default="test",
+        metadata={
+            "help": "The name of the evaluation data set split to use (via the datasets library). Defaults to 'test'"
+        },
+    )
+    audio_column_name: str = field(
+        default="audio",
+        metadata={
+            "help": "The name of the dataset column containing the audio data. Defaults to 'audio'"
+        },
+    )
+    text_column_name: str = field(
+        default="text",
+        metadata={
+            "help": "The name of the dataset column containing the text data. Defaults to 'text'"
+        },
+    )
+    overwrite_cache: bool = field(
+        default=False,
+        metadata={"help": "Overwrite the cached preprocessed datasets or not."},
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of training examples to this "
+                "value if set."
+            )
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of validation examples to this "
+                "value if set."
+            )
+        },
+    )
+    chars_to_ignore: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to remove from the transcripts."},
+    )
+    eval_metrics: List[str] = list_field(
+        default=["wer"],
+        metadata={
+            "help": "A list of metrics the model should be evaluated on. E.g. `'wer cer'`"
+        },
+    )
+    max_duration_in_seconds: float = field(
+        default=20.0,
+        metadata={
+            "help": (
+                "Filter audio files that are longer than `max_duration_in_seconds` seconds to"
+                " 'max_duration_in_seconds`"
+            )
+        },
+    )
+    min_duration_in_seconds: float = field(
+        default=0.0,
+        metadata={
+            "help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"
+        },
+    )
+    preprocessing_only: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Whether to only do data preprocessing and skip training. This is especially useful when data"
+                " preprocessing errors out in distributed training due to timeout. In this case, one should run the"
+                " preprocessing in a non-distributed setup with `preprocessing_only=True` so that the cached datasets"
+                " can consequently be loaded in distributed training"
+            )
+        },
+    )
+    use_auth_token: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "If :obj:`True`, will use the token generated when running"
+                ":obj:`huggingface-cli login` as HTTP bearer authorization for remote files."
+            )
+        },
+    )
+    unk_token: str = field(
+        default="[UNK]",
+        metadata={"help": "The unk token for the tokenizer"},
+    )
+    pad_token: str = field(
+        default="[PAD]",
+        metadata={"help": "The padding token for the tokenizer"},
+    )
+    word_delimiter_token: str = field(
+        default="|",
+        metadata={"help": "The word delimiter token for the tokenizer"},
+    )
+    phoneme_language: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": (
+                "The target language that should be used be"
+                " passed to the tokenizer for tokenization. Note that"
+                " this is only relevant if the model classifies the"
+                " input audio to a sequence of phoneme sequences."
+            )
+        },
+    )
+
+
+@dataclass
+class DataCollatorCTCWithPadding:
+    """
+    Data collator that will dynamically pad the inputs received.
+    Args:
+        processor (:class:`~transformers.AutoProcessor`)
+            The processor used for proccessing the data.
+        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+              sequence if provided).
+            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
+              maximum acceptable input length for the model if that argument is not provided.
+            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+              different lengths).
+        max_length (:obj:`int`, `optional`):
+            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
+        max_length_labels (:obj:`int`, `optional`):
+            Maximum length of the ``labels`` returned list and optionally padding length (see above).
+        pad_to_multiple_of (:obj:`int`, `optional`):
+            If set will pad the sequence to a multiple of the provided value.
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+    """
+
+    processor: AutoProcessor
+    padding: Union[bool, str] = "longest"
+    pad_to_multiple_of: Optional[int] = None
+    pad_to_multiple_of_labels: Optional[int] = None
+
+    def __call__(
+        self, features: List[Dict[str, Union[List[int], torch.Tensor]]]
+    ) -> Dict[str, torch.Tensor]:
+        # split inputs and labels since they have to be of different lenghts and need
+        # different padding methods
+        input_features = [
+            {"input_values": feature["input_values"]} for feature in features
+        ]
+        label_features = [{"input_ids": feature["labels"]} for feature in features]
+
+        batch = self.processor.pad(
+            input_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors="pt",
+        )
+
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
+
+        # replace padding with -100 to ignore loss correctly
+        labels = labels_batch["input_ids"].masked_fill(
+            labels_batch.attention_mask.ne(1), -100
+        )
+
+        batch["labels"] = labels
+        if "attention_mask" in batch:
+            batch["attention_mask"] = batch["attention_mask"].to(torch.long)
+
+        return batch
+
+
+def create_vocabulary_from_data(
+    datasets: DatasetDict,
+    word_delimiter_token: Optional[str] = None,
+    unk_token: Optional[str] = None,
+    pad_token: Optional[str] = None,
+):
+    # Given training and test labels create vocabulary
+    def extract_all_chars(batch):
+        all_text = " ".join(batch["target_text"])
+        # phonemes are split by whitespace
+        vocab = list(set(all_text.split())) + [" "]
+        return {"vocab": [vocab], "all_text": [all_text]}
+
+    vocabs = datasets.map(
+        extract_all_chars,
+        batched=True,
+        batch_size=-1,
+        keep_in_memory=True,
+        remove_columns=datasets["train"].column_names,
+    )
+
+    # take union of all unique characters in each dataset
+    vocab_set = functools.reduce(
+        lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]),
+        vocabs.values(),
+    )
+
+    vocab_dict = {v: k for k, v in enumerate(sorted(list(vocab_set)))}
+
+    # replace white space with delimiter token
+    if word_delimiter_token is not None:
+        vocab_dict[word_delimiter_token] = vocab_dict[" "]
+        del vocab_dict[" "]
+
+    # add unk and pad token
+    if unk_token is not None:
+        vocab_dict[unk_token] = len(vocab_dict)
+
+    if pad_token is not None:
+        vocab_dict[pad_token] = len(vocab_dict)
+
+    return vocab_dict
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser(
+        (ModelArguments, DataTrainingArguments, TrainingArguments)
+    )
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(
+            json_file=os.path.abspath(sys.argv[1])
+        )
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_speech_recognition_ctc", model_args, data_args)
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if (
+        os.path.isdir(training_args.output_dir)
+        and training_args.do_train
+        and not training_args.overwrite_output_dir
+    ):
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    logger.setLevel(
+        logging.INFO if is_main_process(training_args.local_rank) else logging.WARN
+    )
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # 1. First, let's load the dataset
+    raw_datasets = load_dataset(
+        data_args.dataset_name,
+        data_args.dataset_config_name,
+        split=data_args.train_split_name,
+    )
+
+    raw_datasets = raw_datasets.train_test_split(test_size=0.15)
+
+    if training_args.do_train:
+        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:
+        if data_args.max_eval_samples is not None:
+            raw_datasets["test"] = raw_datasets["test"].select(
+                range(data_args.max_eval_samples)
+            )
+
+    # 2. We remove some special characters from the datasets
+    # that make training complicated and do not help in transcribing the speech
+    # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
+    # that could be easily picked up by the model
+    chars_to_ignore_regex = (
+        f'[{"".join(data_args.chars_to_ignore)}]'
+        if data_args.chars_to_ignore is not None
+        else None
+    )
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = (
+                re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+            )
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(
+        desc="dataset map special characters removal"
+    ):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                try:
+                    os.remove(vocab_file)
+                except OSError:
+                    # in shared file-systems it might be the case that
+                    # two processes try to delete the vocab file at the some time
+                    pass
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                vocab_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type
+            if config.tokenizer_class is None
+            else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+    )
+
+    # 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,
+    )
+
+    # freeze encoder
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
+
+    # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
+    # Thankfully, `datasets` takes care of automatically loading and resampling the audio,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = (
+        next(iter(raw_datasets.values()))
+        .features[data_args.audio_column_name]
+        .sampling_rate
+    )
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name,
+            datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate),
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = (
+        data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    )
+    min_input_length = (
+        data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    )
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
+    phoneme_language = data_args.phoneme_language
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(
+            sample["array"], sampling_rate=sample["sampling_rate"]
+        )
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        additional_kwargs = {}
+        if phoneme_language is not None:
+            additional_kwargs["phonemizer_lang"] = phoneme_language
+
+        batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: evaluate.load(metric) for metric in data_args.eval_metrics}
+
+    # for large datasets it is advised to run the preprocessing on a
+    # single machine first with ``args.preprocessing_only`` since there will mostly likely
+    # be a timeout when running the script in distributed mode.
+    # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
+    # cached dataset
+    if data_args.preprocessing_only:
+        logger.info(
+            f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}"
+        )
+        return
+
+    def compute_metrics(pred):
+        pred_logits = pred.predictions
+        pred_ids = np.argmax(pred_logits, axis=-1)
+
+        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
+
+        pred_str = tokenizer.batch_decode(pred_ids)
+        # we do not want to group tokens when computing the metrics
+        label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
+
+        metrics = {
+            k: v.compute(predictions=pred_str, references=label_str)
+            for k, v in eval_metrics.items()
+        }
+
+        return metrics
+
+    # Now save everything to be able to create a single processor later
+    if is_main_process(training_args.local_rank):
+        # save feature extractor, tokenizer and config
+        feature_extractor.save_pretrained(training_args.output_dir)
+        tokenizer.save_pretrained(training_args.output_dir)
+        config.save_pretrained(training_args.output_dir)
+
+    try:
+        processor = AutoProcessor.from_pretrained(training_args.output_dir)
+    except (OSError, KeyError):
+        warnings.warn(
+            "Loading a processor from a feature extractor config that does not"
+            " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
+            " attribute to your `preprocessor_config.json` file to suppress this warning: "
+            " `'processor_class': 'Wav2Vec2Processor'`",
+            FutureWarning,
+        )
+        processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
+
+    # Instantiate custom data collator
+    data_collator = DataCollatorCTCWithPadding(processor=processor)
+
+    # Initialize Trainer
+    trainer = Trainer(
+        model=model,
+        data_collator=data_collator,
+        args=training_args,
+        compute_metrics=compute_metrics,
+        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=vectorized_datasets["test"] 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["test"])
+        )
+        metrics["eval_samples"] = min(
+            max_eval_samples, len(vectorized_datasets["test"])
+        )
+
+        trainer.log_metrics("test", metrics)
+        trainer.save_metrics("test", metrics)
+
+    # Write model card and (optionally) push to hub
+    config_name = (
+        data_args.dataset_config_name
+        if data_args.dataset_config_name is not None
+        else "na"
+    )
+    kwargs = {
+        "finetuned_from": model_args.model_name_or_path,
+        "tasks": "automatic-speech-recognition",
+        "tags": ["automatic-speech-recognition", data_args.dataset_name],
+        "dataset_args": (
+            f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split:"
+            f" {data_args.eval_split_name}"
+        ),
+        "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
+    }
+    if "common_voice" in data_args.dataset_name:
+        kwargs["language"] = config_name
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+    return results
+
+
+if __name__ == "__main__":
+    main()

runs/Jan09_00-52-51_bookbot-pt-2/1673227740.802469/events.out.tfevents.1673227740.bookbot-pt-2.11380.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20aa80f9730f7c36363b97b9caec4240908ec23edcbd227b7f63a990d349d2c0
+size 129

runs/Jan09_00-52-51_bookbot-pt-2/events.out.tfevents.1673227740.bookbot-pt-2.11380.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e104f3a7aae76e3e406df1e908549beab2cf785f91c81343eb46c25099655186
+size 130

runs/Jan09_00-52-51_bookbot-pt-2/events.out.tfevents.1673247932.bookbot-pt-2.11380.2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed25fc9e7b5efd3663ecfa809f39c37a83df79b503f987ed5713e30160e842c4
+size 128

special_tokens_map.json

@@ -0,0 +1,22 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "[PAD]",
+  "unk_token": "[UNK]"
+}

test_results.json

@@ -0,0 +1,10 @@
+{
+    "epoch": 30.0,
+    "eval_cer": 0.0058143100249896515,
+    "eval_loss": 0.06832413375377655,
+    "eval_runtime": 87.0708,
+    "eval_samples": 1965,
+    "eval_samples_per_second": 22.568,
+    "eval_steps_per_second": 2.825,
+    "eval_wer": 0.009874807524938073
+}

tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token": "<s>",
+  "do_lower_case": false,
+  "eos_token": "</s>",
+  "model_max_length": 1000000000000000019884624838656,
+  "name_or_path": "./wav2vec2-ljspeech-gruut",
+  "pad_token": "[PAD]",
+  "replace_word_delimiter_char": " ",
+  "special_tokens_map_file": null,
+  "tokenizer_class": "Wav2Vec2CTCTokenizer",
+  "unk_token": "[UNK]",
+  "word_delimiter_token": "|"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 30.0,
+    "train_loss": 0.15939651108792915,
+    "train_runtime": 19958.548,
+    "train_samples": 11135,
+    "train_samples_per_second": 16.737,
+    "train_steps_per_second": 0.523
+}

trainer_state.json

@@ -0,0 +1,445 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 30.0,
+  "global_step": 10440,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 1.0,
+      "eval_cer": 1.0,
+      "eval_loss": 2.2818498611450195,
+      "eval_runtime": 85.7153,
+      "eval_samples_per_second": 22.925,
+      "eval_steps_per_second": 2.87,
+      "eval_wer": 1.0,
+      "step": 348
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 5e-05,
+      "loss": 2.6692,
+      "step": 500
+    },
+    {
+      "epoch": 2.0,
+      "eval_cer": 0.029949254143222897,
+      "eval_loss": 0.2044876664876938,
+      "eval_runtime": 85.7259,
+      "eval_samples_per_second": 22.922,
+      "eval_steps_per_second": 2.87,
+      "eval_wer": 0.052662850639351944,
+      "step": 696
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 0.0001,
+      "loss": 0.2225,
+      "step": 1000
+    },
+    {
+      "epoch": 3.0,
+      "eval_cer": 0.018884051083140417,
+      "eval_loss": 0.11616221815347672,
+      "eval_runtime": 85.4863,
+      "eval_samples_per_second": 22.986,
+      "eval_steps_per_second": 2.878,
+      "eval_wer": 0.031858806989355296,
+      "step": 1044
+    },
+    {
+      "epoch": 4.0,
+      "eval_cer": 0.014687169423704908,
+      "eval_loss": 0.09268919378519058,
+      "eval_runtime": 85.2634,
+      "eval_samples_per_second": 23.046,
+      "eval_steps_per_second": 2.885,
+      "eval_wer": 0.023540536921737965,
+      "step": 1392
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 9.470338983050848e-05,
+      "loss": 0.0868,
+      "step": 1500
+    },
+    {
+      "epoch": 5.0,
+      "eval_cer": 0.014284728716635749,
+      "eval_loss": 0.07971413433551788,
+      "eval_runtime": 85.4679,
+      "eval_samples_per_second": 22.991,
+      "eval_steps_per_second": 2.878,
+      "eval_wer": 0.021841735288210484,
+      "step": 1740
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 8.940677966101694e-05,
+      "loss": 0.0598,
+      "step": 2000
+    },
+    {
+      "epoch": 6.0,
+      "eval_cer": 0.012847440477103041,
+      "eval_loss": 0.07152710855007172,
+      "eval_runtime": 85.5884,
+      "eval_samples_per_second": 22.959,
+      "eval_steps_per_second": 2.874,
+      "eval_wer": 0.01967429872129611,
+      "step": 2088
+    },
+    {
+      "epoch": 7.0,
+      "eval_cer": 0.010252656108666656,
+      "eval_loss": 0.06518065184354782,
+      "eval_runtime": 85.121,
+      "eval_samples_per_second": 23.085,
+      "eval_steps_per_second": 2.89,
+      "eval_wer": 0.015975430139921,
+      "step": 2436
+    },
+    {
+      "epoch": 7.18,
+      "learning_rate": 8.411016949152542e-05,
+      "loss": 0.0447,
+      "step": 2500
+    },
+    {
+      "epoch": 8.0,
+      "eval_cer": 0.00946693853772211,
+      "eval_loss": 0.057057663798332214,
+      "eval_runtime": 84.3173,
+      "eval_samples_per_second": 23.305,
+      "eval_steps_per_second": 2.918,
+      "eval_wer": 0.015230635335073977,
+      "step": 2784
+    },
+    {
+      "epoch": 8.62,
+      "learning_rate": 7.88135593220339e-05,
+      "loss": 0.0368,
+      "step": 3000
+    },
+    {
+      "epoch": 9.0,
+      "eval_cer": 0.011180186119245098,
+      "eval_loss": 0.060811206698417664,
+      "eval_runtime": 84.3248,
+      "eval_samples_per_second": 23.303,
+      "eval_steps_per_second": 2.917,
+      "eval_wer": 0.01630180089710116,
+      "step": 3132
+    },
+    {
+      "epoch": 10.0,
+      "eval_cer": 0.008297944102902173,
+      "eval_loss": 0.058583296835422516,
+      "eval_runtime": 84.6721,
+      "eval_samples_per_second": 23.207,
+      "eval_steps_per_second": 2.905,
+      "eval_wer": 0.013657360915846555,
+      "step": 3480
+    },
+    {
+      "epoch": 10.06,
+      "learning_rate": 7.351694915254238e-05,
+      "loss": 0.0303,
+      "step": 3500
+    },
+    {
+      "epoch": 11.0,
+      "eval_cer": 0.008535575758504913,
+      "eval_loss": 0.06412886828184128,
+      "eval_runtime": 87.3267,
+      "eval_samples_per_second": 22.502,
+      "eval_steps_per_second": 2.817,
+      "eval_wer": 0.014125995849233446,
+      "step": 3828
+    },
+    {
+      "epoch": 11.49,
+      "learning_rate": 6.822033898305085e-05,
+      "loss": 0.0273,
+      "step": 4000
+    },
+    {
+      "epoch": 12.0,
+      "eval_cer": 0.007933831082220552,
+      "eval_loss": 0.06564020365476608,
+      "eval_runtime": 84.2667,
+      "eval_samples_per_second": 23.319,
+      "eval_steps_per_second": 2.919,
+      "eval_wer": 0.013071567249112941,
+      "step": 4176
+    },
+    {
+      "epoch": 12.93,
+      "learning_rate": 6.29343220338983e-05,
+      "loss": 0.0232,
+      "step": 4500
+    },
+    {
+      "epoch": 13.0,
+      "eval_cer": 0.008225121498765848,
+      "eval_loss": 0.06898853182792664,
+      "eval_runtime": 84.1359,
+      "eval_samples_per_second": 23.355,
+      "eval_steps_per_second": 2.924,
+      "eval_wer": 0.0132640423110397,
+      "step": 4524
+    },
+    {
+      "epoch": 14.0,
+      "eval_cer": 0.00787250678400049,
+      "eval_loss": 0.05983823910355568,
+      "eval_runtime": 84.1873,
+      "eval_samples_per_second": 23.341,
+      "eval_steps_per_second": 2.922,
+      "eval_wer": 0.012803775858606146,
+      "step": 4872
+    },
+    {
+      "epoch": 14.37,
+      "learning_rate": 5.763771186440679e-05,
+      "loss": 0.0189,
+      "step": 5000
+    },
+    {
+      "epoch": 15.0,
+      "eval_cer": 0.007420240084627531,
+      "eval_loss": 0.06711488217115402,
+      "eval_runtime": 84.8039,
+      "eval_samples_per_second": 23.171,
+      "eval_steps_per_second": 2.901,
+      "eval_wer": 0.012100823458525808,
+      "step": 5220
+    },
+    {
+      "epoch": 15.8,
+      "learning_rate": 5.2341101694915265e-05,
+      "loss": 0.017,
+      "step": 5500
+    },
+    {
+      "epoch": 16.0,
+      "eval_cer": 0.006906649087034511,
+      "eval_loss": 0.06541039049625397,
+      "eval_runtime": 84.2563,
+      "eval_samples_per_second": 23.322,
+      "eval_steps_per_second": 2.92,
+      "eval_wer": 0.011364397134632121,
+      "step": 5568
+    },
+    {
+      "epoch": 17.0,
+      "eval_cer": 0.007335919174574946,
+      "eval_loss": 0.07511687278747559,
+      "eval_runtime": 84.0992,
+      "eval_samples_per_second": 23.365,
+      "eval_steps_per_second": 2.925,
+      "eval_wer": 0.011807926625159,
+      "step": 5916
+    },
+    {
+      "epoch": 17.24,
+      "learning_rate": 4.705508474576271e-05,
+      "loss": 0.0146,
+      "step": 6000
+    },
+    {
+      "epoch": 18.0,
+      "eval_cer": 0.006753338341484355,
+      "eval_loss": 0.06527850776910782,
+      "eval_runtime": 83.7988,
+      "eval_samples_per_second": 23.449,
+      "eval_steps_per_second": 2.936,
+      "eval_wer": 0.011171922072705363,
+      "step": 6264
+    },
+    {
+      "epoch": 18.68,
+      "learning_rate": 4.175847457627119e-05,
+      "loss": 0.0127,
+      "step": 6500
+    },
+    {
+      "epoch": 19.0,
+      "eval_cer": 0.006921980161589526,
+      "eval_loss": 0.06817645579576492,
+      "eval_runtime": 84.0515,
+      "eval_samples_per_second": 23.379,
+      "eval_steps_per_second": 2.927,
+      "eval_wer": 0.01123886992033206,
+      "step": 6612
+    },
+    {
+      "epoch": 20.0,
+      "eval_cer": 0.006814662639704417,
+      "eval_loss": 0.06784532964229584,
+      "eval_runtime": 83.9653,
+      "eval_samples_per_second": 23.403,
+      "eval_steps_per_second": 2.93,
+      "eval_wer": 0.01137276561558546,
+      "step": 6960
+    },
+    {
+      "epoch": 20.11,
+      "learning_rate": 3.6461864406779664e-05,
+      "loss": 0.0114,
+      "step": 7000
+    },
+    {
+      "epoch": 21.0,
+      "eval_cer": 0.006584696521379184,
+      "eval_loss": 0.06555593758821487,
+      "eval_runtime": 83.8204,
+      "eval_samples_per_second": 23.443,
+      "eval_steps_per_second": 2.935,
+      "eval_wer": 0.011113342706032,
+      "step": 7308
+    },
+    {
+      "epoch": 21.55,
+      "learning_rate": 3.117584745762712e-05,
+      "loss": 0.0101,
+      "step": 7500
+    },
+    {
+      "epoch": 22.0,
+      "eval_cer": 0.006596194827295445,
+      "eval_loss": 0.06685744225978851,
+      "eval_runtime": 84.0101,
+      "eval_samples_per_second": 23.39,
+      "eval_steps_per_second": 2.928,
+      "eval_wer": 0.010920867644105242,
+      "step": 7656
+    },
+    {
+      "epoch": 22.99,
+      "learning_rate": 2.5879237288135593e-05,
+      "loss": 0.0092,
+      "step": 8000
+    },
+    {
+      "epoch": 23.0,
+      "eval_cer": 0.006477378999494075,
+      "eval_loss": 0.06765928864479065,
+      "eval_runtime": 84.2885,
+      "eval_samples_per_second": 23.313,
+      "eval_steps_per_second": 2.919,
+      "eval_wer": 0.010778603467898508,
+      "step": 8004
+    },
+    {
+      "epoch": 24.0,
+      "eval_cer": 0.006331733791221427,
+      "eval_loss": 0.0652570053935051,
+      "eval_runtime": 84.1568,
+      "eval_samples_per_second": 23.349,
+      "eval_steps_per_second": 2.923,
+      "eval_wer": 0.010402021824998326,
+      "step": 8352
+    },
+    {
+      "epoch": 24.43,
+      "learning_rate": 2.058262711864407e-05,
+      "loss": 0.0088,
+      "step": 8500
+    },
+    {
+      "epoch": 25.0,
+      "eval_cer": 0.006266576724362611,
+      "eval_loss": 0.0673212930560112,
+      "eval_runtime": 83.9435,
+      "eval_samples_per_second": 23.409,
+      "eval_steps_per_second": 2.931,
+      "eval_wer": 0.01020117828211823,
+      "step": 8700
+    },
+    {
+      "epoch": 25.86,
+      "learning_rate": 1.5286016949152543e-05,
+      "loss": 0.0074,
+      "step": 9000
+    },
+    {
+      "epoch": 26.0,
+      "eval_cer": 0.006350897634415196,
+      "eval_loss": 0.06691750884056091,
+      "eval_runtime": 84.013,
+      "eval_samples_per_second": 23.389,
+      "eval_steps_per_second": 2.928,
+      "eval_wer": 0.0104857066345317,
+      "step": 9048
+    },
+    {
+      "epoch": 27.0,
+      "eval_cer": 0.006113265978812455,
+      "eval_loss": 0.0707407295703888,
+      "eval_runtime": 84.4435,
+      "eval_samples_per_second": 23.27,
+      "eval_steps_per_second": 2.913,
+      "eval_wer": 0.01013423043449153,
+      "step": 9396
+    },
+    {
+      "epoch": 27.3,
+      "learning_rate": 9.989406779661017e-06,
+      "loss": 0.0066,
+      "step": 9500
+    },
+    {
+      "epoch": 28.0,
+      "eval_cer": 0.0059829518450948225,
+      "eval_loss": 0.06726762652397156,
+      "eval_runtime": 84.279,
+      "eval_samples_per_second": 23.315,
+      "eval_steps_per_second": 2.919,
+      "eval_wer": 0.009966860815424784,
+      "step": 9744
+    },
+    {
+      "epoch": 28.74,
+      "learning_rate": 4.692796610169492e-06,
+      "loss": 0.0058,
+      "step": 10000
+    },
+    {
+      "epoch": 29.0,
+      "eval_cer": 0.005867968785932206,
+      "eval_loss": 0.06885003298521042,
+      "eval_runtime": 84.0405,
+      "eval_samples_per_second": 23.382,
+      "eval_steps_per_second": 2.927,
+      "eval_wer": 0.010000334739238134,
+      "step": 10092
+    },
+    {
+      "epoch": 30.0,
+      "eval_cer": 0.0058143100249896515,
+      "eval_loss": 0.06832413375377655,
+      "eval_runtime": 84.1565,
+      "eval_samples_per_second": 23.349,
+      "eval_steps_per_second": 2.923,
+      "eval_wer": 0.009874807524938073,
+      "step": 10440
+    },
+    {
+      "epoch": 30.0,
+      "step": 10440,
+      "total_flos": 2.004174730615405e+19,
+      "train_loss": 0.15939651108792915,
+      "train_runtime": 19958.548,
+      "train_samples_per_second": 16.737,
+      "train_steps_per_second": 0.523
+    }
+  ],
+  "max_steps": 10440,
+  "num_train_epochs": 30,
+  "total_flos": 2.004174730615405e+19,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de95cb920d4b0059afc57ae8c102afd51b2c3c85288d1d6f137d0d604bc5c375
+size 129

vocab.json

@@ -0,0 +1,45 @@
+{
+  "[PAD]": 42,
+  "[UNK]": 41,
+  "aɪ": 1,
+  "aʊ": 2,
+  "b": 3,
+  "d": 4,
+  "d͡ʒ": 5,
+  "eɪ": 6,
+  "f": 7,
+  "h": 8,
+  "i": 9,
+  "j": 10,
+  "k": 11,
+  "l": 12,
+  "m": 13,
+  "n": 14,
+  "oʊ": 15,
+  "p": 16,
+  "s": 17,
+  "t": 18,
+  "t͡ʃ": 19,
+  "u": 20,
+  "v": 21,
+  "w": 22,
+  "z": 23,
+  "|": 0,
+  "æ": 24,
+  "ð": 25,
+  "ŋ": 26,
+  "ɑ": 27,
+  "ɔ": 28,
+  "ɔɪ": 29,
+  "ə": 30,
+  "ɚ": 31,
+  "ɛ": 32,
+  "ɡ": 33,
+  "ɪ": 34,
+  "ɹ": 35,
+  "ʃ": 36,
+  "ʊ": 37,
+  "ʌ": 38,
+  "ʒ": 39,
+  "θ": 40
+}