initial commit

.gitattributes

@@ -25,3 +25,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+language_model/text_for_lm.txt filter=lfs diff=lfs merge=lfs -text
+language_model/5gram.arpa filter=lfs diff=lfs merge=lfs -text
+language_model/5gram_correct.arpa filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,5 @@
+checkpoint-*/
+
+.ipynb_checkpoints/
+
+wandb

README.md

@@ -0,0 +1,75 @@
+---
+language:
+- fr
+tags:
+- automatic-speech-recognition
+- mozilla-foundation/common_voice_7_0
+- generated_from_trainer
+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 [./checkpoint-6000](https://huggingface.co/./checkpoint-6000) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - FR dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.2619
+- Wer: 0.2457
+
+## 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: 16
+- eval_batch_size: 16
+- seed: 42
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 128
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- lr_scheduler_warmup_steps: 2000
+- num_epochs: 2.0
+- mixed_precision_training: Native AMP
+
+### Training results
+
+| Training Loss | Epoch | Step | Validation Loss | Wer    |
+|:-------------:|:-----:|:----:|:---------------:|:------:|
+| 3.495         | 0.16  | 500  | 3.3883          | 1.0    |
+| 2.9095        | 0.32  | 1000 | 2.9152          | 1.0000 |
+| 1.8434        | 0.49  | 1500 | 1.0473          | 0.7446 |
+| 1.4298        | 0.65  | 2000 | 0.5729          | 0.5130 |
+| 1.1937        | 0.81  | 2500 | 0.3795          | 0.3450 |
+| 1.1248        | 0.97  | 3000 | 0.3321          | 0.3052 |
+| 1.0835        | 1.13  | 3500 | 0.3038          | 0.2805 |
+| 1.0479        | 1.3   | 4000 | 0.2910          | 0.2689 |
+| 1.0413        | 1.46  | 4500 | 0.2798          | 0.2593 |
+| 1.014         | 1.62  | 5000 | 0.2727          | 0.2512 |
+| 1.004         | 1.78  | 5500 | 0.2646          | 0.2471 |
+| 0.9949        | 1.94  | 6000 | 0.2619          | 0.2457 |
+
+
+### Framework versions
+
+- Transformers 4.17.0.dev0
+- Pytorch 1.10.2+cu102
+- Datasets 1.18.2.dev0
+- Tokenizers 0.11.0

all_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 2.0,
+    "eval_loss": 0.26187804341316223,
+    "eval_runtime": 789.1417,
+    "eval_samples": 15941,
+    "eval_samples_per_second": 20.2,
+    "eval_steps_per_second": 1.263,
+    "eval_wer": 0.24574541380398318,
+    "train_loss": 0.02791261456650878,
+    "train_runtime": 12943.901,
+    "train_samples": 395042,
+    "train_samples_per_second": 61.039,
+    "train_steps_per_second": 0.477
+}

config.json

@@ -0,0 +1,107 @@
+{
+  "_name_or_path": "./checkpoint-6000",
+  "activation_dropout": 0.1,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.0,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.0,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.25,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.75,
+  "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": 45,
+  "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": "4.17.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 46,
+  "xvector_output_dim": 512
+}

create_lm_model.ipynb

@@ -0,0 +1,320 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "04c8de09",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from datasets import load_dataset\n",
+    "import re"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 23,
+   "id": "1eae750a",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Reusing dataset common_voice (/workspace/.cache/huggingface/datasets/mozilla-foundation___common_voice/fr/7.0.0/fe20cac47c166e25b1f096ab661832e3da7cf298ed4a91dcaa1343ad972d175b)\n"
+     ]
+    }
+   ],
+   "source": [
+    "dataset = load_dataset(\"mozilla-foundation/common_voice_7_0\", \"fr\", split=\"train\", use_auth_token=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "id": "da1cfcaa",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "c110c54654c045b9a2cbc6cad43fa685",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "0ex [00:00, ?ex/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "chars_to_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\\'’ ]'\n",
+    "\n",
+    "def extract_text(batch):\n",
+    "    batch[\"text\"] = re.sub(chars_to_ignore_regex, \"\", batch[\"sentence\"].lower()).replace('’', \"'\")\n",
+    "    return batch\n",
+    "\n",
+    "dataset = dataset.map(extract_text, remove_columns=[\"sentence\"])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 25,
+   "id": "bb306916",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "d21bc14560b747f49105f598a2ffe2ff",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Pushing dataset shards to the dataset hub:   0%|          | 0/29 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "dataset.push_to_hub(f\"common_voice_7_0_fr_processed\", split=\"train\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 26,
+   "id": "312d9c63",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with open(\"text_for_lm.txt\", \"w\") as file:\n",
+    "    file.write(\" \".join(dataset[\"text\"]))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 29,
+   "id": "b605f94f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "sys.path.append(\"test.arpa\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 33,
+   "id": "9262368d",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Loading the LM will be faster if you build a binary file.\n",
+      "Reading /workspace/xls-r-300m-fr/language_model/5gram.arpa\n",
+      "----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100\n"
+     ]
+    },
+    {
+     "ename": "OSError",
+     "evalue": "Cannot read model 'language_model/5gram.arpa' (End of file Byte: 0)",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
+      "File \u001b[0;32mkenlm.pyx:139\u001b[0m, in \u001b[0;36mkenlm.Model.__init__\u001b[0;34m()\u001b[0m\n",
+      "\u001b[0;31mRuntimeError\u001b[0m: End of file Byte: 0",
+      "\nThe above exception was the direct cause of the following exception:\n",
+      "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
+      "Input \u001b[0;32mIn [33]\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01msys\u001b[39;00m\n\u001b[1;32m      5\u001b[0m LM \u001b[38;5;241m=\u001b[39m os\u001b[38;5;241m.\u001b[39mpath\u001b[38;5;241m.\u001b[39mjoin(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mlanguage_model/\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m'\u001b[39m\u001b[38;5;124m5gram.arpa\u001b[39m\u001b[38;5;124m'\u001b[39m)\n\u001b[0;32m----> 6\u001b[0m model \u001b[38;5;241m=\u001b[39m \u001b[43mkenlm\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mLanguageModel\u001b[49m\u001b[43m(\u001b[49m\u001b[43mLM\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m      7\u001b[0m \u001b[38;5;28mprint\u001b[39m(\u001b[38;5;124m'\u001b[39m\u001b[38;5;132;01m{0}\u001b[39;00m\u001b[38;5;124m-gram model\u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;241m.\u001b[39mformat(model\u001b[38;5;241m.\u001b[39morder))\n",
+      "File \u001b[0;32mkenlm.pyx:142\u001b[0m, in \u001b[0;36mkenlm.Model.__init__\u001b[0;34m()\u001b[0m\n",
+      "\u001b[0;31mOSError\u001b[0m: Cannot read model 'language_model/5gram.arpa' (End of file Byte: 0)"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "import kenlm\n",
+    "import sys\n",
+    "\n",
+    "LM = os.path.join(\"language_model/\", '5gram.arpa')\n",
+    "model = kenlm.LanguageModel(LM)\n",
+    "print('{0}-gram model'.format(model.order))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 38,
+   "id": "130f7f47",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "14200cf29dc74389aeceb56701ee9a5f",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/255 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "9da79daa4d6b4a6694b96d0608d5b6ec",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/221 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "862bc04ec9e241608eab1aa2e21ef41f",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/730 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "ada7a9b9364644658c3e8b58c1a34d2c",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/85.0 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "e7cbe24e2bec4efa997a0e7d3d87d082",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/683 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "6b903049438d4067a0ec13d55483e9e4",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/227 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "9d59cebc3b7b46228c365f6b73cee4ee",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/78.0 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "7a2709adf3764eb29750700b6f58ec40",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/1.50G [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "7276b3af03ce408ebcb485a0ff5641a7",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading:   0%|          | 0.00/11.1M [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "from transformers import Wav2Vec2ProcessorWithLM\n",
+    "\n",
+    "processor = Wav2Vec2ProcessorWithLM.from_pretrained(\"Harveenchadha/hindi_model_with_lm_vakyansh\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e75ab227",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.8.8"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

eval.py

@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline
+
+
+def log_results(result: Dataset, args: Dict[str, str]):
+    """DO NOT CHANGE. This function computes and logs the result metrics."""
+
+    log_outputs = args.log_outputs
+    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
+
+    # load metric
+    wer = load_metric("wer")
+    cer = load_metric("cer")
+
+    # compute metrics
+    wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
+    cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
+
+    # print & log results
+    result_str = f"WER: {wer_result}\n" f"CER: {cer_result}"
+    print(result_str)
+
+    with open(f"{dataset_id}_eval_results.txt", "w") as f:
+        f.write(result_str)
+
+    # log all results in text file. Possibly interesting for analysis
+    if log_outputs is not None:
+        pred_file = f"log_{dataset_id}_predictions.txt"
+        target_file = f"log_{dataset_id}_targets.txt"
+
+        with open(pred_file, "w") as p, open(target_file, "w") as t:
+
+            # mapping function to write output
+            def write_to_file(batch, i):
+                p.write(f"{i}" + "\n")
+                p.write(batch["prediction"] + "\n")
+                t.write(f"{i}" + "\n")
+                t.write(batch["target"] + "\n")
+
+            result.map(write_to_file, with_indices=True)
+
+
+def normalize_text(text: str) -> str:
+    """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
+
+    chars_to_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\'’ ]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    text = re.sub(chars_to_ignore_regex, "", text.lower()).replace('’', "'")
+
+    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
+    # note that order is important here!
+    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]
+
+    for t in token_sequences_to_ignore:
+        text = " ".join(text.split(t))
+
+    return text
+
+
+def main(args):
+    # load dataset
+    dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
+
+    # for testing: only process the first two examples as a test
+    # dataset = dataset.select(range(10))
+
+    # load processor
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    asr = pipeline("automatic-speech-recognition", model=args.model_id)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        prediction = asr(
+            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
+        )
+
+        batch["prediction"] = prediction["text"]
+        batch["target"] = normalize_text(batch["sentence"])
+        return batch
+
+    # run inference on all examples
+    result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
+
+    # compute and log_results
+    # do not change function below
+    log_results(result, args)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
+    )
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        required=True,
+        help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
+    )
+    parser.add_argument(
+        "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'`  for Common Voice"
+    )
+    parser.add_argument("--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`")
+    parser.add_argument(
+        "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
+    )
+    parser.add_argument(
+        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
+    )
+    parser.add_argument(
+        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
+    )
+    args = parser.parse_args()
+
+    main(args)

eval_lm.py

@@ -0,0 +1,131 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline, Wav2Vec2ProcessorWithLM
+
+
+def log_results(result: Dataset, args: Dict[str, str]):
+    """DO NOT CHANGE. This function computes and logs the result metrics."""
+
+    log_outputs = args.log_outputs
+    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
+
+    # load metric
+    wer = load_metric("wer")
+    cer = load_metric("cer")
+
+    # compute metrics
+    wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
+    cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
+
+    # print & log results
+    result_str = f"WER: {wer_result}\n" f"CER: {cer_result}"
+    print(result_str)
+
+    with open(f"{dataset_id}_eval_results.txt", "w") as f:
+        f.write(result_str)
+
+    # log all results in text file. Possibly interesting for analysis
+    if log_outputs is not None:
+        pred_file = f"log_{dataset_id}_predictions.txt"
+        target_file = f"log_{dataset_id}_targets.txt"
+
+        with open(pred_file, "w") as p, open(target_file, "w") as t:
+
+            # mapping function to write output
+            def write_to_file(batch, i):
+                p.write(f"{i}" + "\n")
+                p.write(batch["prediction"] + "\n")
+                t.write(f"{i}" + "\n")
+                t.write(batch["target"] + "\n")
+
+            result.map(write_to_file, with_indices=True)
+
+
+def normalize_text(text: str) -> str:
+    """DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""
+
+    chars_to_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\'’ ]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    text = re.sub(chars_to_ignore_regex, "", text.lower()).replace('’', "'")
+
+    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
+    # note that order is important here!
+    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]
+
+    for t in token_sequences_to_ignore:
+        text = " ".join(text.split(t))
+
+    return text
+
+
+def main(args):
+    # load dataset
+    dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
+
+    # for testing: only process the first two examples as a test
+    # dataset = dataset.select(range(10))
+
+    # load processor
+    processor = Wav2Vec2ProcessorWithLM.from_pretrained("Plim/")
+
+    model = Wav2Vec2ForCTC.from_pretrained(model_id)
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    asr = pipeline("automatic-speech-recognition", model=args.model_id)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        prediction = asr(
+            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
+        )
+
+        batch["prediction"] = prediction["text"]
+        batch["target"] = normalize_text(batch["sentence"])
+        return batch
+
+    # run inference on all examples
+    result = dataset.map(map_to_pred, remove_columns=dataset.column_names)
+
+    # compute and log_results
+    # do not change function below
+    log_results(result, args)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--model_id", type=str, required=True, help="Model identifier. Should be loadable with 🤗 Transformers"
+    )
+    parser.add_argument(
+        "--dataset",
+        type=str,
+        required=True,
+        help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
+    )
+    parser.add_argument(
+        "--config", type=str, required=True, help="Config of the dataset. *E.g.* `'en'`  for Common Voice"
+    )
+    parser.add_argument("--split", type=str, required=True, help="Split of the dataset. *E.g.* `'test'`")
+    parser.add_argument(
+        "--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to 5 seconds."
+    )
+    parser.add_argument(
+        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
+    )
+    parser.add_argument(
+        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
+    )
+    args = parser.parse_args()
+
+    main(args)

eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 2.0,
+    "eval_loss": 0.26187804341316223,
+    "eval_runtime": 789.1417,
+    "eval_samples": 15941,
+    "eval_samples_per_second": 20.2,
+    "eval_steps_per_second": 1.263,
+    "eval_wer": 0.24574541380398318
+}

language_model/5gram.arpa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adf4dc592314c578e7a8c2c0b4618fdef0a6a6ac06c4dd75acb30108c4d8d133
+size 507942255

language_model/5gram_correct.arpa

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1909f82cc26181421c79ba6b774a59e68cbb80f6c61e53e4d6f68d09d6d987de
+size 507942072

language_model/text_for_lm.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1140ec9580ad66c6aface6216aef499a8febf71f3f4fc6073cc8c5115069241
+size 23840468

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:ed573364e574cdd8727dcc0047037d7add9029bcf15566f97cd6f972f1fc93ee
+size 1262112241

run.sh

@@ -0,0 +1,37 @@
+export WANDB_PROJECT="xls-r-300-fr"
+python run_speech_recognition_ctc.py \
+    --activation_dropout="0.1" \
+    --dataset_name="mozilla-foundation/common_voice_7_0" \
+    --dataset_config_name="fr" \
+    --eval_steps="1000" \
+    --evaluation_strategy="steps" \
+    --feat_proj_dropout="0.0" \
+    --freeze_feature_encoder \
+    --fp16 \
+    --gradient_accumulation_steps="8" \
+    --gradient_checkpointing \
+    --group_by_length \
+    --layerdrop="0.0" \
+    --learning_rate="7.5e-5" \
+    --length_column_name="input_length" \
+    --load_best_model_at_end \
+    --logging_steps="100" \
+    --mask_feature_length="64" \
+    --mask_feature_prob="0.25" \
+    --mask_time_length="10" \
+    --mask_time_prob="0.75" \
+    --model_name_or_path="./checkpoint-6000" \
+    --num_train_epochs="6.0" \
+    --output_dir="./" \
+    --overwrite_output_dir \
+    --per_device_train_batch_size="16" \
+    --per_device_eval_batch_size="16" \
+    --preprocessing_num_workers="4" \
+    --push_to_hub \
+    --report_to="wandb" \
+    --save_steps="1000" \
+    --save_total_limit="3" \
+    --text_column_name="sentence" \
+    --use_auth_token \
+    --warmup_steps="2000" \
+    --do_train --do_eval

run_speech_recognition_ctc.py

@@ -0,0 +1,734 @@
+#!/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("4.16.0.dev0")
+
+require_version("datasets>=1.13.3", "To fix: pip install -r examples/pytorch/text-classification/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 vector"
+            "span 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'"
+        },
+    )
+    eval_split_name: str = field(
+        default="test",
+        metadata={
+            "help": "The name of the training 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:`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_ignore_regex = '[^a-zàâäçéèêëîïôöùûüÿ\'’ ]'
+    text_column_name = data_args.text_column_name
+
+    def remove_and_replace_special_characters(batch):
+        batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name].lower()).replace('’', "'") + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_and_replace_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,
+            "eos_token": None,
+            "bos_token": None,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "feat_proj_dropout": model_args.feat_proj_dropout,
+            "attention_dropout": model_args.attention_dropout,
+            "hidden_dropout": model_args.hidden_dropout,
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "activation_dropout": model_args.activation_dropout,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+    )
+
+    # freeze encoder
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
+
+    # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
+    # Thankfully, `datasets` takes care of automatically loading and resampling the audio,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
+    phoneme_language = data_args.phoneme_language
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        additional_kwargs = {}
+        if phoneme_language is not None:
+            additional_kwargs["phonemizer_lang"] = phoneme_language
+
+        batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: 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: {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()

special_tokens_map.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "pad_token": "[PAD]"}

test_results/mozilla-foundation_common_voice_7_0_fr_test_eval_results.txt

@@ -0,0 +1,2 @@
+WER: 0.24561764914155493
+CER: 0.07285207821118034

tokenizer_config.json

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

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 2.0,
+    "train_loss": 0.02791261456650878,
+    "train_runtime": 12943.901,
+    "train_samples": 395042,
+    "train_samples_per_second": 61.039,
+    "train_steps_per_second": 0.477
+}

train_results/all_results.json

@@ -0,0 +1,14 @@
+{
+    "epoch": 2.0,
+    "eval_loss": 0.26187804341316223,
+    "eval_runtime": 722.142,
+    "eval_samples": 15941,
+    "eval_samples_per_second": 22.075,
+    "eval_steps_per_second": 1.381,
+    "eval_wer": 0.24574541380398318,
+    "train_loss": 1.788894365302016,
+    "train_runtime": 52105.5599,
+    "train_samples": 395042,
+    "train_samples_per_second": 15.163,
+    "train_steps_per_second": 0.118
+}

train_results/eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 2.0,
+    "eval_loss": 0.26187804341316223,
+    "eval_runtime": 722.142,
+    "eval_samples": 15941,
+    "eval_samples_per_second": 22.075,
+    "eval_steps_per_second": 1.381,
+    "eval_wer": 0.24574541380398318
+}

train_results/train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 2.0,
+    "train_loss": 1.788894365302016,
+    "train_runtime": 52105.5599,
+    "train_samples": 395042,
+    "train_samples_per_second": 15.163,
+    "train_steps_per_second": 0.118
+}

train_results/trainer_state.json

@@ -0,0 +1,499 @@
+{
+  "best_metric": 0.26187804341316223,
+  "best_model_checkpoint": "./checkpoint-6000",
+  "epoch": 1.9998784982382245,
+  "global_step": 6172,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.7499999999999997e-06,
+      "loss": 12.1043,
+      "step": 100
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.499999999999999e-06,
+      "loss": 6.4771,
+      "step": 200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.1249999999999999e-05,
+      "loss": 4.4866,
+      "step": 300
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.4999999999999999e-05,
+      "loss": 3.8842,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.8712499999999997e-05,
+      "loss": 3.495,
+      "step": 500
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 3.3882696628570557,
+      "eval_runtime": 721.337,
+      "eval_samples_per_second": 22.099,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 2.2462499999999997e-05,
+      "loss": 3.171,
+      "step": 600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 2.6212499999999997e-05,
+      "loss": 3.0275,
+      "step": 700
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.99625e-05,
+      "loss": 2.9681,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 3.37125e-05,
+      "loss": 2.9347,
+      "step": 900
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.7462499999999996e-05,
+      "loss": 2.9095,
+      "step": 1000
+    },
+    {
+      "epoch": 0.32,
+      "eval_loss": 2.9152133464813232,
+      "eval_runtime": 718.1623,
+      "eval_samples_per_second": 22.197,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.9999871219487068,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 4.12125e-05,
+      "loss": 2.8888,
+      "step": 1100
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 4.4962499999999995e-05,
+      "loss": 2.8347,
+      "step": 1200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 4.871249999999999e-05,
+      "loss": 2.5318,
+      "step": 1300
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.2462499999999994e-05,
+      "loss": 2.0502,
+      "step": 1400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.62125e-05,
+      "loss": 1.8434,
+      "step": 1500
+    },
+    {
+      "epoch": 0.49,
+      "eval_loss": 1.0473320484161377,
+      "eval_runtime": 720.1235,
+      "eval_samples_per_second": 22.136,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7446153648029981,
+      "step": 1500
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5.9962499999999994e-05,
+      "loss": 1.7339,
+      "step": 1600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 6.367499999999999e-05,
+      "loss": 1.6535,
+      "step": 1700
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 6.7425e-05,
+      "loss": 1.5793,
+      "step": 1800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 7.1175e-05,
+      "loss": 1.5056,
+      "step": 1900
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 7.492499999999999e-05,
+      "loss": 1.4298,
+      "step": 2000
+    },
+    {
+      "epoch": 0.65,
+      "eval_loss": 0.5728740692138672,
+      "eval_runtime": 712.5783,
+      "eval_samples_per_second": 22.371,
+      "eval_steps_per_second": 1.399,
+      "eval_wer": 0.5129521000882147,
+      "step": 2000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 7.325623202301054e-05,
+      "loss": 1.3592,
+      "step": 2100
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 7.145853307766058e-05,
+      "loss": 1.2917,
+      "step": 2200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 6.966083413231063e-05,
+      "loss": 1.2536,
+      "step": 2300
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.788111217641419e-05,
+      "loss": 1.2345,
+      "step": 2400
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 6.608341323106423e-05,
+      "loss": 1.1937,
+      "step": 2500
+    },
+    {
+      "epoch": 0.81,
+      "eval_loss": 0.3795304000377655,
+      "eval_runtime": 716.4435,
+      "eval_samples_per_second": 22.25,
+      "eval_steps_per_second": 1.392,
+      "eval_wer": 0.34504806732645216,
+      "step": 2500
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 6.428571428571427e-05,
+      "loss": 1.1806,
+      "step": 2600
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.248801534036433e-05,
+      "loss": 1.1651,
+      "step": 2700
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 6.069031639501438e-05,
+      "loss": 1.1455,
+      "step": 2800
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.889261744966442e-05,
+      "loss": 1.1312,
+      "step": 2900
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5.709491850431447e-05,
+      "loss": 1.1248,
+      "step": 3000
+    },
+    {
+      "epoch": 0.97,
+      "eval_loss": 0.3320523500442505,
+      "eval_runtime": 716.2808,
+      "eval_samples_per_second": 22.255,
+      "eval_steps_per_second": 1.392,
+      "eval_wer": 0.30515830344552264,
+      "step": 3000
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.5297219558964525e-05,
+      "loss": 1.1017,
+      "step": 3100
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5.3499520613614567e-05,
+      "loss": 1.0978,
+      "step": 3200
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5.1701821668264615e-05,
+      "loss": 1.0954,
+      "step": 3300
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 4.990412272291467e-05,
+      "loss": 1.0867,
+      "step": 3400
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 4.812440076701821e-05,
+      "loss": 1.0835,
+      "step": 3500
+    },
+    {
+      "epoch": 1.13,
+      "eval_loss": 0.3037940561771393,
+      "eval_runtime": 714.1597,
+      "eval_samples_per_second": 22.321,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.2805032742445413,
+      "step": 3500
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 4.632670182166826e-05,
+      "loss": 1.0808,
+      "step": 3600
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 4.4529002876318304e-05,
+      "loss": 1.0648,
+      "step": 3700
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.273130393096836e-05,
+      "loss": 1.0541,
+      "step": 3800
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.093360498561841e-05,
+      "loss": 1.0621,
+      "step": 3900
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 3.913590604026845e-05,
+      "loss": 1.0479,
+      "step": 4000
+    },
+    {
+      "epoch": 1.3,
+      "eval_loss": 0.2910499572753906,
+      "eval_runtime": 718.1665,
+      "eval_samples_per_second": 22.197,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.26888727197800427,
+      "step": 4000
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 3.7338207094918506e-05,
+      "loss": 1.0428,
+      "step": 4100
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 3.555848513902205e-05,
+      "loss": 1.047,
+      "step": 4200
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 3.37607861936721e-05,
+      "loss": 1.0397,
+      "step": 4300
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 3.1963087248322145e-05,
+      "loss": 1.0347,
+      "step": 4400
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 3.0165388302972194e-05,
+      "loss": 1.0413,
+      "step": 4500
+    },
+    {
+      "epoch": 1.46,
+      "eval_loss": 0.27976545691490173,
+      "eval_runtime": 713.7382,
+      "eval_samples_per_second": 22.335,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.2592995627901586,
+      "step": 4500
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 2.836768935762224e-05,
+      "loss": 1.0238,
+      "step": 4600
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 2.656999041227229e-05,
+      "loss": 1.0269,
+      "step": 4700
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 2.4772291466922337e-05,
+      "loss": 1.021,
+      "step": 4800
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 2.2974592521572386e-05,
+      "loss": 1.0186,
+      "step": 4900
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 2.1176893576222434e-05,
+      "loss": 1.014,
+      "step": 5000
+    },
+    {
+      "epoch": 1.62,
+      "eval_loss": 0.27265554666519165,
+      "eval_runtime": 707.3075,
+      "eval_samples_per_second": 22.538,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.25117351242409997,
+      "step": 5000
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.9379194630872483e-05,
+      "loss": 1.0074,
+      "step": 5100
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1.759947267497603e-05,
+      "loss": 1.0193,
+      "step": 5200
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1.5801773729626078e-05,
+      "loss": 1.0044,
+      "step": 5300
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.4004074784276125e-05,
+      "loss": 1.0005,
+      "step": 5400
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.2206375838926173e-05,
+      "loss": 1.004,
+      "step": 5500
+    },
+    {
+      "epoch": 1.78,
+      "eval_loss": 0.26460376381874084,
+      "eval_runtime": 719.6956,
+      "eval_samples_per_second": 22.15,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.2470782921128375,
+      "step": 5500
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1.0408676893576222e-05,
+      "loss": 1.0048,
+      "step": 5600
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 8.610977948226269e-06,
+      "loss": 0.9988,
+      "step": 5700
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 6.813279002876318e-06,
+      "loss": 0.9919,
+      "step": 5800
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 5.015580057526366e-06,
+      "loss": 0.9886,
+      "step": 5900
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 3.217881112176414e-06,
+      "loss": 0.9949,
+      "step": 6000
+    },
+    {
+      "epoch": 1.94,
+      "eval_loss": 0.26187804341316223,
+      "eval_runtime": 717.4473,
+      "eval_samples_per_second": 22.219,
+      "eval_steps_per_second": 1.39,
+      "eval_wer": 0.24574541380398318,
+      "step": 6000
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1.4201821668264622e-06,
+      "loss": 0.9931,
+      "step": 6100
+    },
+    {
+      "epoch": 2.0,
+      "step": 6172,
+      "total_flos": 1.1573983785360925e+20,
+      "train_loss": 1.788894365302016,
+      "train_runtime": 52105.5599,
+      "train_samples_per_second": 15.163,
+      "train_steps_per_second": 0.118
+    }
+  ],
+  "max_steps": 6172,
+  "num_train_epochs": 2,
+  "total_flos": 1.1573983785360925e+20,
+  "trial_name": null,
+  "trial_params": null
+}

trainer_state.json

@@ -0,0 +1,499 @@
+{
+  "best_metric": 0.26187804341316223,
+  "best_model_checkpoint": "./checkpoint-6000",
+  "epoch": 1.9998784982382245,
+  "global_step": 6172,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.7499999999999997e-06,
+      "loss": 12.1043,
+      "step": 100
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 7.499999999999999e-06,
+      "loss": 6.4771,
+      "step": 200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.1249999999999999e-05,
+      "loss": 4.4866,
+      "step": 300
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.4999999999999999e-05,
+      "loss": 3.8842,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.8712499999999997e-05,
+      "loss": 3.495,
+      "step": 500
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 3.3882696628570557,
+      "eval_runtime": 721.337,
+      "eval_samples_per_second": 22.099,
+      "eval_steps_per_second": 1.382,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 2.2462499999999997e-05,
+      "loss": 3.171,
+      "step": 600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 2.6212499999999997e-05,
+      "loss": 3.0275,
+      "step": 700
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.99625e-05,
+      "loss": 2.9681,
+      "step": 800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 3.37125e-05,
+      "loss": 2.9347,
+      "step": 900
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 3.7462499999999996e-05,
+      "loss": 2.9095,
+      "step": 1000
+    },
+    {
+      "epoch": 0.32,
+      "eval_loss": 2.9152133464813232,
+      "eval_runtime": 718.1623,
+      "eval_samples_per_second": 22.197,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.9999871219487068,
+      "step": 1000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 4.12125e-05,
+      "loss": 2.8888,
+      "step": 1100
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 4.4962499999999995e-05,
+      "loss": 2.8347,
+      "step": 1200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 4.871249999999999e-05,
+      "loss": 2.5318,
+      "step": 1300
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.2462499999999994e-05,
+      "loss": 2.0502,
+      "step": 1400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.62125e-05,
+      "loss": 1.8434,
+      "step": 1500
+    },
+    {
+      "epoch": 0.49,
+      "eval_loss": 1.0473320484161377,
+      "eval_runtime": 720.1235,
+      "eval_samples_per_second": 22.136,
+      "eval_steps_per_second": 1.384,
+      "eval_wer": 0.7446153648029981,
+      "step": 1500
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5.9962499999999994e-05,
+      "loss": 1.7339,
+      "step": 1600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 6.367499999999999e-05,
+      "loss": 1.6535,
+      "step": 1700
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 6.7425e-05,
+      "loss": 1.5793,
+      "step": 1800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 7.1175e-05,
+      "loss": 1.5056,
+      "step": 1900
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 7.492499999999999e-05,
+      "loss": 1.4298,
+      "step": 2000
+    },
+    {
+      "epoch": 0.65,
+      "eval_loss": 0.5728740692138672,
+      "eval_runtime": 712.5783,
+      "eval_samples_per_second": 22.371,
+      "eval_steps_per_second": 1.399,
+      "eval_wer": 0.5129521000882147,
+      "step": 2000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 7.325623202301054e-05,
+      "loss": 1.3592,
+      "step": 2100
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 7.145853307766058e-05,
+      "loss": 1.2917,
+      "step": 2200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 6.966083413231063e-05,
+      "loss": 1.2536,
+      "step": 2300
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 6.788111217641419e-05,
+      "loss": 1.2345,
+      "step": 2400
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 6.608341323106423e-05,
+      "loss": 1.1937,
+      "step": 2500
+    },
+    {
+      "epoch": 0.81,
+      "eval_loss": 0.3795304000377655,
+      "eval_runtime": 716.4435,
+      "eval_samples_per_second": 22.25,
+      "eval_steps_per_second": 1.392,
+      "eval_wer": 0.34504806732645216,
+      "step": 2500
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 6.428571428571427e-05,
+      "loss": 1.1806,
+      "step": 2600
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 6.248801534036433e-05,
+      "loss": 1.1651,
+      "step": 2700
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 6.069031639501438e-05,
+      "loss": 1.1455,
+      "step": 2800
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5.889261744966442e-05,
+      "loss": 1.1312,
+      "step": 2900
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5.709491850431447e-05,
+      "loss": 1.1248,
+      "step": 3000
+    },
+    {
+      "epoch": 0.97,
+      "eval_loss": 0.3320523500442505,
+      "eval_runtime": 716.2808,
+      "eval_samples_per_second": 22.255,
+      "eval_steps_per_second": 1.392,
+      "eval_wer": 0.30515830344552264,
+      "step": 3000
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5.5297219558964525e-05,
+      "loss": 1.1017,
+      "step": 3100
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5.3499520613614567e-05,
+      "loss": 1.0978,
+      "step": 3200
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5.1701821668264615e-05,
+      "loss": 1.0954,
+      "step": 3300
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 4.990412272291467e-05,
+      "loss": 1.0867,
+      "step": 3400
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 4.812440076701821e-05,
+      "loss": 1.0835,
+      "step": 3500
+    },
+    {
+      "epoch": 1.13,
+      "eval_loss": 0.3037940561771393,
+      "eval_runtime": 714.1597,
+      "eval_samples_per_second": 22.321,
+      "eval_steps_per_second": 1.396,
+      "eval_wer": 0.2805032742445413,
+      "step": 3500
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 4.632670182166826e-05,
+      "loss": 1.0808,
+      "step": 3600
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 4.4529002876318304e-05,
+      "loss": 1.0648,
+      "step": 3700
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.273130393096836e-05,
+      "loss": 1.0541,
+      "step": 3800
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.093360498561841e-05,
+      "loss": 1.0621,
+      "step": 3900
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 3.913590604026845e-05,
+      "loss": 1.0479,
+      "step": 4000
+    },
+    {
+      "epoch": 1.3,
+      "eval_loss": 0.2910499572753906,
+      "eval_runtime": 718.1665,
+      "eval_samples_per_second": 22.197,
+      "eval_steps_per_second": 1.388,
+      "eval_wer": 0.26888727197800427,
+      "step": 4000
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 3.7338207094918506e-05,
+      "loss": 1.0428,
+      "step": 4100
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 3.555848513902205e-05,
+      "loss": 1.047,
+      "step": 4200
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 3.37607861936721e-05,
+      "loss": 1.0397,
+      "step": 4300
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 3.1963087248322145e-05,
+      "loss": 1.0347,
+      "step": 4400
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 3.0165388302972194e-05,
+      "loss": 1.0413,
+      "step": 4500
+    },
+    {
+      "epoch": 1.46,
+      "eval_loss": 0.27976545691490173,
+      "eval_runtime": 713.7382,
+      "eval_samples_per_second": 22.335,
+      "eval_steps_per_second": 1.397,
+      "eval_wer": 0.2592995627901586,
+      "step": 4500
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 2.836768935762224e-05,
+      "loss": 1.0238,
+      "step": 4600
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 2.656999041227229e-05,
+      "loss": 1.0269,
+      "step": 4700
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 2.4772291466922337e-05,
+      "loss": 1.021,
+      "step": 4800
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 2.2974592521572386e-05,
+      "loss": 1.0186,
+      "step": 4900
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 2.1176893576222434e-05,
+      "loss": 1.014,
+      "step": 5000
+    },
+    {
+      "epoch": 1.62,
+      "eval_loss": 0.27265554666519165,
+      "eval_runtime": 707.3075,
+      "eval_samples_per_second": 22.538,
+      "eval_steps_per_second": 1.41,
+      "eval_wer": 0.25117351242409997,
+      "step": 5000
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.9379194630872483e-05,
+      "loss": 1.0074,
+      "step": 5100
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1.759947267497603e-05,
+      "loss": 1.0193,
+      "step": 5200
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1.5801773729626078e-05,
+      "loss": 1.0044,
+      "step": 5300
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.4004074784276125e-05,
+      "loss": 1.0005,
+      "step": 5400
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.2206375838926173e-05,
+      "loss": 1.004,
+      "step": 5500
+    },
+    {
+      "epoch": 1.78,
+      "eval_loss": 0.26460376381874084,
+      "eval_runtime": 719.6956,
+      "eval_samples_per_second": 22.15,
+      "eval_steps_per_second": 1.385,
+      "eval_wer": 0.2470782921128375,
+      "step": 5500
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1.0408676893576222e-05,
+      "loss": 1.0048,
+      "step": 5600
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 8.610977948226269e-06,
+      "loss": 0.9988,
+      "step": 5700
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 6.813279002876318e-06,
+      "loss": 0.9919,
+      "step": 5800
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 5.015580057526366e-06,
+      "loss": 0.9886,
+      "step": 5900
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 3.217881112176414e-06,
+      "loss": 0.9949,
+      "step": 6000
+    },
+    {
+      "epoch": 1.94,
+      "eval_loss": 0.26187804341316223,
+      "eval_runtime": 717.4473,
+      "eval_samples_per_second": 22.219,
+      "eval_steps_per_second": 1.39,
+      "eval_wer": 0.24574541380398318,
+      "step": 6000
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1.4201821668264622e-06,
+      "loss": 0.9931,
+      "step": 6100
+    },
+    {
+      "epoch": 2.0,
+      "step": 6172,
+      "total_flos": 1.1573983785360925e+20,
+      "train_loss": 0.02791261456650878,
+      "train_runtime": 12943.901,
+      "train_samples_per_second": 61.039,
+      "train_steps_per_second": 0.477
+    }
+  ],
+  "max_steps": 6172,
+  "num_train_epochs": 2,
+  "total_flos": 1.1573983785360925e+20,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf1c15b9bf70ef516fdb9c5d070521f8cd9c035b43933970064d3a0f192e8f80
+size 2991

vocab.json

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