Training in progress, step 500

.gitignore

@@ -0,0 +1 @@
+checkpoint-*/

added_tokens.json

@@ -0,0 +1 @@
+{"<s>": 59, "</s>": 60}

config.json

@@ -0,0 +1,107 @@
+{
+  "_name_or_path": "facebook/wav2vec2-xls-r-300m",
+  "activation_dropout": 0.1,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.0,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.0,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.0,
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.25,
+  "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": 58,
+  "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": 61,
+  "xvector_output_dim": 512
+}

eval.py

@@ -0,0 +1,137 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+import torch
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline
+
+
+def log_results(result: Dataset, args: Dict[str, str]):
+    """DO NOT CHANGE. This function computes and logs the result metrics."""
+
+    log_outputs = args.log_outputs
+    dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])
+
+    # load metric
+    wer = load_metric("wer")
+    cer = load_metric("cer")
+
+    # compute metrics
+    wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
+    cer_result = cer.compute(references=result["target"], predictions=result["prediction"])
+
+    # print & log results
+    result_str = f"WER: {wer_result}\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 = '[,?.!\-\;\:"“%‘”�—’…–]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    text = re.sub(chars_to_ignore_regex, "", text.lower())
+
+    # In addition, we can normalize the target text, e.g. removing new lines characters etc...
+    # note that order is important here!
+    token_sequences_to_ignore = ["\n\n", "\n", "   ", "  "]
+
+    for t in token_sequences_to_ignore:
+        text = " ".join(text.split(t))
+
+    return text
+
+
+def main(args):
+    # load dataset
+    dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)
+
+    # for testing: only process the first two examples as a test
+    # dataset = dataset.select(range(10))
+
+    # load processor
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    if args.device is None:
+        args.device = 0 if torch.cuda.is_available() else -1
+    asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        prediction = asr(
+            batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s
+        )
+
+        batch["prediction"] = prediction["text"]
+        batch["target"] = normalize_text(batch["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."
+    )
+    parser.add_argument(
+        "--device",
+        type=int,
+        default=None,
+        help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
+    )
+    args = parser.parse_args()
+
+    main(args)

nohup.out

@@ -0,0 +1,550 @@
+02/02/2022 18:04:15 - WARNING - __main__ - Process rank: -1, device: cuda:0, n_gpu: 1distributed training: False, 16-bits training: True
+02/02/2022 18:04:15 - INFO - __main__ - Training/evaluation parameters TrainingArguments(
+_n_gpu=1,
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+bf16=False,
+bf16_full_eval=False,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_pin_memory=True,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+debug=[],
+deepspeed=None,
+disable_tqdm=False,
+do_eval=True,
+do_predict=False,
+do_train=True,
+eval_accumulation_steps=None,
+eval_steps=500,
+evaluation_strategy=IntervalStrategy.STEPS,
+fp16=True,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+gradient_accumulation_steps=4,
+gradient_checkpointing=True,
+greater_is_better=None,
+group_by_length=True,
+half_precision_backend=auto,
+hub_model_id=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=7.5e-05,
+length_column_name=input_length,
+load_best_model_at_end=False,
+local_rank=-1,
+log_level=-1,
+log_level_replica=-1,
+log_on_each_node=True,
+logging_dir=./runs/Feb02_18-04-15_job-86e1d453-0156-4b77-a98d-7d457c737175,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=100,
+logging_strategy=IntervalStrategy.STEPS,
+lr_scheduler_type=SchedulerType.LINEAR,
+max_grad_norm=1.0,
+max_steps=-1,
+metric_for_best_model=None,
+mp_parameters=,
+no_cuda=False,
+num_train_epochs=50.0,
+optim=OptimizerNames.ADAMW_HF,
+output_dir=./,
+overwrite_output_dir=True,
+past_index=-1,
+per_device_eval_batch_size=8,
+per_device_train_batch_size=8,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+remove_unused_columns=True,
+report_to=[],
+resume_from_checkpoint=None,
+run_name=./,
+save_on_each_node=False,
+save_steps=500,
+save_strategy=IntervalStrategy.STEPS,
+save_total_limit=3,
+seed=42,
+sharded_ddp=[],
+skip_memory_metrics=True,
+tf32=None,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_legacy_prediction_loop=False,
+warmup_ratio=0.0,
+warmup_steps=2000,
+weight_decay=0.0,
+xpu_backend=None,
+)
+02/02/2022 18:04:18 - WARNING - datasets.builder - Reusing dataset common_voice (/workspace/.cache/huggingface/datasets/mozilla-foundation___common_voice/ur/8.0.0/b8bc4d453193c06a43269b46cd87f075c70f152ac963b7f28f7a2760c45ec3e8)
+02/02/2022 18:04:20 - WARNING - datasets.builder - Reusing dataset common_voice (/workspace/.cache/huggingface/datasets/mozilla-foundation___common_voice/ur/8.0.0/b8bc4d453193c06a43269b46cd87f075c70f152ac963b7f28f7a2760c45ec3e8)
+02/02/2022 18:04:20 - WARNING - datasets.arrow_dataset - Loading cached processed dataset at /workspace/.cache/huggingface/datasets/mozilla-foundation___common_voice/ur/8.0.0/b8bc4d453193c06a43269b46cd87f075c70f152ac963b7f28f7a2760c45ec3e8/cache-eefb1dcecdbc6361.arrow
+02/02/2022 18:04:20 - WARNING - datasets.arrow_dataset - Loading cached processed dataset at /workspace/.cache/huggingface/datasets/mozilla-foundation___common_voice/ur/8.0.0/b8bc4d453193c06a43269b46cd87f075c70f152ac963b7f28f7a2760c45ec3e8/cache-bebf53ae59038f0e.arrow
+loading configuration file https://huggingface.co/facebook/wav2vec2-xls-r-300m/resolve/main/config.json from cache at /workspace/.cache/huggingface/transformers/dabc27df63e37bd2a7a221c7774e35f36a280fbdf917cf54cadfc7df8c786f6f.a3e4c3c967d9985881e0ae550a5f6f668f897db5ab2e0802f9b97973b15970e6
+Model config Wav2Vec2Config {
+  "_name_or_path": "facebook/wav2vec2-xls-r-300m",
+  "activation_dropout": 0.0,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForPreTraining"
+  ],
+  "attention_dropout": 0.1,
+  "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": "sum",
+  "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": 0.1,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "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_prob": 0.075,
+  "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": 0,
+  "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": 32,
+  "xvector_output_dim": 512
+}
+
+
  0%|          | 0/1 [00:00<?, ?ba/s]
+
  0%|          | 0/1 [00:00<?, ?ba/s]
+Didn't find file ./tokenizer.json. We won't load it.
+loading file ./vocab.json
+loading file ./tokenizer_config.json
+loading file ./added_tokens.json
+loading file ./special_tokens_map.json
+loading file None
+Adding <s> to the vocabulary
+Adding </s> to the vocabulary
+loading configuration file https://huggingface.co/facebook/wav2vec2-xls-r-300m/resolve/main/config.json from cache at /workspace/.cache/huggingface/transformers/dabc27df63e37bd2a7a221c7774e35f36a280fbdf917cf54cadfc7df8c786f6f.a3e4c3c967d9985881e0ae550a5f6f668f897db5ab2e0802f9b97973b15970e6
+Model config Wav2Vec2Config {
+  "_name_or_path": "facebook/wav2vec2-xls-r-300m",
+  "activation_dropout": 0.0,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForPreTraining"
+  ],
+  "attention_dropout": 0.1,
+  "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": "sum",
+  "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": 0.1,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "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_prob": 0.075,
+  "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": 0,
+  "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": 32,
+  "xvector_output_dim": 512
+}
+
+loading feature extractor configuration file https://huggingface.co/facebook/wav2vec2-xls-r-300m/resolve/main/preprocessor_config.json from cache at /workspace/.cache/huggingface/transformers/6fb028b95b394059e7d3b367bbca2382b576c66aebe896f04d2cd34e1b575f5b.d4484dc1c81456a2461485e7168b04347a7b9a4e3b1ef3aba723323b33e12326
+Feature extractor Wav2Vec2FeatureExtractor {
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}
+
+loading weights file https://huggingface.co/facebook/wav2vec2-xls-r-300m/resolve/main/pytorch_model.bin from cache at /workspace/.cache/huggingface/transformers/1e6a6507f3b689035cd4b247e2a37c154e27f39143f31357a49b4e38baeccc36.1edb32803799e27ed554eb7dd935f6745b1a0b17b0ea256442fe24db6eb546cd
+Some weights of the model checkpoint at facebook/wav2vec2-xls-r-300m were not used when initializing Wav2Vec2ForCTC: ['quantizer.weight_proj.weight', 'quantizer.codevectors', 'quantizer.weight_proj.bias', 'project_q.bias', 'project_q.weight', 'project_hid.bias', 'project_hid.weight']
+- This IS expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
+- This IS NOT expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
+Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at facebook/wav2vec2-xls-r-300m and are newly initialized: ['lm_head.weight', 'lm_head.bias']
+You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
+
+
+
  0%|          | 0/1 [00:00<?, ?ba/s]
+
  0%|          | 0/1 [00:00<?, ?ba/s]
+Configuration saved in ./preprocessor_config.json
+tokenizer config file saved in ./tokenizer_config.json
+Special tokens file saved in ./special_tokens_map.json
+added tokens file saved in ./added_tokens.json
+Configuration saved in ./config.json
+loading feature extractor configuration file ./preprocessor_config.json
+loading configuration file ./config.json
+Model config Wav2Vec2Config {
+  "_name_or_path": "./",
+  "activation_dropout": 0.1,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForPreTraining"
+  ],
+  "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": 58,
+  "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": 61,
+  "xvector_output_dim": 512
+}
+
+loading feature extractor configuration file ./preprocessor_config.json
+Feature extractor Wav2Vec2FeatureExtractor {
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}
+
+Didn't find file ./tokenizer.json. We won't load it.
+loading file ./vocab.json
+loading file ./tokenizer_config.json
+loading file ./added_tokens.json
+loading file ./special_tokens_map.json
+loading file None
+Adding <s> to the vocabulary
+Adding </s> to the vocabulary
+/workspace/xls-r-300m-ur/./ is already a clone of https://huggingface.co/HarrisDePerceptron/xls-r-300m-ur. Make sure you pull the latest changes with `repo.git_pull()`.
+02/02/2022 18:04:42 - WARNING - huggingface_hub.repository - /workspace/xls-r-300m-ur/./ is already a clone of https://huggingface.co/HarrisDePerceptron/xls-r-300m-ur. Make sure you pull the latest changes with `repo.git_pull()`.
+Using amp half precision backend
+The following columns in the training set  don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.
+/opt/conda/lib/python3.8/site-packages/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use thePyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning
+  warnings.warn(
+***** Running training *****
+  Num examples = 810
+  Num Epochs = 50
+  Instantaneous batch size per device = 8
+  Total train batch size (w. parallel, distributed & accumulation) = 32
+  Gradient Accumulation steps = 4
+  Total optimization steps = 1250
+
  0%|          | 0/1250 [00:00<?, ?it/s]
  0%|          | 1/1250 [00:02<53:39,  2.58s/it]
  0%|          | 2/1250 [00:04<43:18,  2.08s/it]
  0%|          | 3/1250 [00:05<37:42,  1.81s/it]
  0%|          | 4/1250 [00:07<34:12,  1.65s/it]
  0%|          | 5/1250 [00:08<31:07,  1.50s/it]
  0%|          | 6/1250 [00:09<27:31,  1.33s/it]
  1%|          | 7/1250 [00:11<35:11,  1.70s/it]
  1%|          | 8/1250 [00:13<35:39,  1.72s/it]
  1%|          | 9/1250 [00:15<34:52,  1.69s/it]
  1%|          | 10/1250 [00:16<33:13,  1.61s/it]
  1%|          | 11/1250 [00:17<30:26,  1.47s/it]
  1%|          | 12/1250 [00:18<27:37,  1.34s/it]
  1%|          | 13/1250 [00:21<34:42,  1.68s/it]
  1%|          | 14/1250 [00:23<34:49,  1.69s/it]
  1%|          | 15/1250 [00:24<34:04,  1.66s/it]
  1%|▏         | 16/1250 [00:26<32:22,  1.57s/it]
  1%|▏         | 17/1250 [00:27<29:57,  1.46s/it]
  1%|▏         | 18/1250 [00:28<26:57,  1.31s/it]
  2%|▏         | 19/1250 [00:30<32:14,  1.57s/it]
  2%|▏         | 20/1250 [00:32<32:52,  1.60s/it]
  2%|▏         | 21/1250 [00:33<32:10,  1.57s/it]
  2%|▏         | 22/1250 [00:34<30:44,  1.50s/it]
  2%|▏         | 23/1250 [00:36<28:38,  1.40s/it]
  2%|▏         | 24/1250 [00:37<26:12,  1.28s/it]
  2%|▏         | 25/1250 [00:38<28:56,  1.42s/it]
  2%|▏         | 26/1250 [00:41<38:28,  1.89s/it]
  2%|▏         | 27/1250 [00:43<37:43,  1.85s/it]
  2%|▏         | 28/1250 [00:45<35:59,  1.77s/it]
  2%|▏         | 29/1250 [00:46<33:21,  1.64s/it]
  2%|▏         | 30/1250 [00:47<30:47,  1.51s/it]
  2%|▏         | 31/1250 [00:48<27:31,  1.36s/it]
  3%|▎         | 32/1250 [00:51<33:43,  1.66s/it]
  3%|▎         | 33/1250 [00:52<34:25,  1.70s/it]
  3%|▎         | 34/1250 [00:54<33:34,  1.66s/it]
  3%|▎         | 35/1250 [00:55<31:55,  1.58s/it]
  3%|▎         | 36/1250 [00:56<29:40,  1.47s/it]
  3%|▎         | 37/1250 [00:58<26:54,  1.33s/it]
  3%|▎         | 38/1250 [01:00<33:02,  1.64s/it]
  3%|▎         | 39/1250 [01:02<33:43,  1.67s/it]
  3%|▎         | 40/1250 [01:03<32:32,  1.61s/it]
  3%|▎         | 41/1250 [01:04<30:50,  1.53s/it]
  3%|▎         | 42/1250 [01:06<28:27,  1.41s/it]
  3%|▎         | 43/1250 [01:07<25:45,  1.28s/it]
  4%|▎         | 44/1250 [01:09<31:31,  1.57s/it]
  4%|▎         | 45/1250 [01:10<32:20,  1.61s/it]
  4%|▎         | 46/1250 [01:12<31:58,  1.59s/it]
  4%|▍         | 47/1250 [01:13<30:42,  1.53s/it]
  4%|▍         | 48/1250 [01:15<28:46,  1.44s/it]
  4%|▍         | 49/1250 [01:16<26:22,  1.32s/it]
  4%|▍         | 50/1250 [01:17<29:12,  1.46s/it]
  4%|▍         | 51/1250 [01:20<37:20,  1.87s/it]
  4%|▍         | 52/1250 [01:22<36:38,  1.83s/it]
  4%|▍         | 53/1250 [01:24<35:02,  1.76s/it]
  4%|▍         | 54/1250 [01:25<32:55,  1.65s/it]
  4%|▍         | 55/1250 [01:26<30:10,  1.51s/it]
  4%|▍         | 56/1250 [01:27<26:59,  1.36s/it]
  5%|▍         | 57/1250 [01:29<32:08,  1.62s/it]
  5%|▍         | 58/1250 [01:31<32:44,  1.65s/it]
  5%|▍         | 59/1250 [01:33<31:55,  1.61s/it]
  5%|▍         | 60/1250 [01:34<30:25,  1.53s/it]
  5%|▍         | 61/1250 [01:35<28:26,  1.44s/it]
  5%|▍         | 62/1250 [01:36<26:14,  1.33s/it]
  5%|▌         | 63/1250 [01:39<32:19,  1.63s/it]
  5%|▌         | 64/1250 [01:40<33:12,  1.68s/it]
  5%|▌         | 65/1250 [01:42<32:10,  1.63s/it]
  5%|▌         | 66/1250 [01:43<30:43,  1.56s/it]
  5%|▌         | 67/1250 [01:45<28:35,  1.45s/it]
  5%|▌         | 68/1250 [01:46<25:57,  1.32s/it]
  6%|▌         | 69/1250 [01:48<33:24,  1.70s/it]
  6%|▌         | 70/1250 [01:50<34:10,  1.74s/it]
  6%|▌         | 71/1250 [01:51<32:41,  1.66s/it]
  6%|▌         | 72/1250 [01:53<30:43,  1.57s/it]
  6%|▌         | 73/1250 [01:54<28:32,  1.46s/it]
  6%|▌         | 74/1250 [01:55<25:40,  1.31s/it]
  6%|▌         | 75/1250 [01:57<28:23,  1.45s/it]
  6%|▌         | 76/1250 [02:00<37:02,  1.89s/it]
  6%|▌         | 77/1250 [02:01<36:23,  1.86s/it]
  6%|▌         | 78/1250 [02:03<34:41,  1.78s/it]
  6%|▋         | 79/1250 [02:04<32:21,  1.66s/it]
  6%|▋         | 80/1250 [02:06<29:42,  1.52s/it]
  6%|▋         | 81/1250 [02:07<26:30,  1.36s/it]
  7%|▋         | 82/1250 [02:09<32:23,  1.66s/it]
  7%|▋         | 83/1250 [02:11<32:49,  1.69s/it]
  7%|▋         | 84/1250 [02:12<31:56,  1.64s/it]
  7%|▋         | 85/1250 [02:14<30:09,  1.55s/it]
  7%|▋         | 86/1250 [02:15<27:51,  1.44s/it]
  7%|▋         | 87/1250 [02:16<25:16,  1.30s/it]
  7%|▋         | 88/1250 [02:18<32:07,  1.66s/it]
  7%|▋         | 89/1250 [02:20<32:48,  1.70s/it]
  7%|▋         | 90/1250 [02:22<31:43,  1.64s/it]
  7%|▋         | 91/1250 [02:23<30:21,  1.57s/it]
  7%|▋         | 92/1250 [02:24<28:26,  1.47s/it]
  7%|▋         | 93/1250 [02:25<26:29,  1.37s/it]
  8%|▊         | 94/1250 [02:28<31:23,  1.63s/it]
  8%|▊         | 95/1250 [02:29<31:51,  1.66s/it]
  8%|▊         | 96/1250 [02:31<30:50,  1.60s/it]
  8%|▊         | 97/1250 [02:32<29:30,  1.54s/it]
  8%|▊         | 98/1250 [02:33<27:25,  1.43s/it]
  8%|▊         | 99/1250 [02:34<24:44,  1.29s/it]
  8%|▊         | 100/1250 [02:36<26:59,  1.41s/it]
                                                  
  8%|▊         | 100/1250 [02:36<26:59,  1.41s/it]
  8%|▊         | 101/1250 [02:39<35:08,  1.84s/it]
  8%|▊         | 102/1250 [02:41<34:28,  1.80s/it]
  8%|▊         | 103/1250 [02:42<32:30,  1.70s/it]
  8%|▊         | 104/1250 [02:43<30:05,  1.58s/it]
  8%|▊         | 105/1250 [02:44<27:37,  1.45s/it]
  8%|▊         | 106/1250 [02:45<24:44,  1.30s/it]
  9%|▊         | 107/1250 [02:48<30:38,  1.61s/it]
  9%|▊         | 108/1250 [02:49<31:24,  1.65s/it]
  9%|▊         | 109/1250 [02:51<30:45,  1.62s/it]
  9%|▉         | 110/1250 [02:52<29:22,  1.55s/it]
  9%|▉         | 111/1250 [02:54<27:20,  1.44s/it]
  9%|▉         | 112/1250 [02:55<24:41,  1.30s/it]
  9%|▉         | 113/1250 [02:57<29:34,  1.56s/it]
  9%|▉         | 114/1250 [02:58<30:11,  1.59s/it]
  9%|▉         | 115/1250 [03:00<29:50,  1.58s/it]
  9%|▉         | 116/1250 [03:01<28:49,  1.53s/it]
  9%|▉         | 117/1250 [03:03<26:52,  1.42s/it]
  9%|▉         | 118/1250 [03:04<24:25,  1.29s/it]
 10%|▉         | 119/1250 [03:06<30:27,  1.62s/it]
 10%|▉         | 120/1250 [03:08<31:40,  1.68s/it]
 10%|▉         | 121/1250 [03:09<30:52,  1.64s/it]
 10%|▉         | 122/1250 [03:11<29:20,  1.56s/it]
 10%|▉         | 123/1250 [03:12<27:11,  1.45s/it]
 10%|▉         | 124/1250 [03:13<24:31,  1.31s/it]
 10%|█         | 125/1250 [03:15<28:39,  1.53s/it]
 10%|█         | 126/1250 [03:18<36:45,  1.96s/it]
 10%|█         | 127/1250 [03:20<35:48,  1.91s/it]
 10%|█         | 128/1250 [03:21<33:48,  1.81s/it]
 10%|█         | 129/1250 [03:23<31:32,  1.69s/it]
 10%|█         | 130/1250 [03:24<28:58,  1.55s/it]
 10%|█         | 131/1250 [03:25<25:53,  1.39s/it]
 11%|█         | 132/1250 [03:27<31:02,  1.67s/it]
 11%|█         | 133/1250 [03:29<31:42,  1.70s/it]
 11%|█         | 134/1250 [03:30<30:44,  1.65s/it]
 11%|█         | 135/1250 [03:32<29:06,  1.57s/it]
 11%|█         | 136/1250 [03:33<26:59,  1.45s/it]
 11%|█         | 137/1250 [03:34<24:33,  1.32s/it]
 11%|█         | 138/1250 [03:36<30:43,  1.66s/it]
 11%|█         | 139/1250 [03:38<31:10,  1.68s/it]
 11%|█         | 140/1250 [03:40<30:34,  1.65s/it]
 11%|█▏        | 141/1250 [03:41<29:08,  1.58s/it]
 11%|█▏        | 142/1250 [03:42<26:41,  1.45s/it]
 11%|█▏        | 143/1250 [03:43<24:14,  1.31s/it]
 12%|█▏        | 144/1250 [03:46<29:19,  1.59s/it]
 12%|█▏        | 145/1250 [03:47<29:55,  1.62s/it]
 12%|█▏        | 146/1250 [03:49<29:20,  1.59s/it]
 12%|█▏        | 147/1250 [03:50<28:09,  1.53s/it]
 12%|█▏        | 148/1250 [03:51<26:42,  1.45s/it]
 12%|█▏        | 149/1250 [03:52<24:23,  1.33s/it]
 12%|█▏        | 150/1250 [03:54<27:24,  1.49s/it]
 12%|█▏        | 151/1250 [03:57<34:21,  1.88s/it]
 12%|█▏        | 152/1250 [03:59<33:23,  1.82s/it]
 12%|█▏        | 153/1250 [04:00<31:32,  1.72s/it]
 12%|█▏        | 154/1250 [04:02<29:24,  1.61s/it]
 12%|█▏        | 155/1250 [04:03<27:13,  1.49s/it]
 12%|█▏        | 156/1250 [04:04<24:24,  1.34s/it]
 13%|█▎        | 157/1250 [04:06<30:38,  1.68s/it]
 13%|█▎        | 158/1250 [04:08<30:45,  1.69s/it]
 13%|█▎        | 159/1250 [04:10<30:02,  1.65s/it]
 13%|█▎        | 160/1250 [04:11<28:40,  1.58s/it]
 13%|█▎        | 161/1250 [04:12<26:29,  1.46s/it]
 13%|█▎        | 162/1250 [04:13<23:57,  1.32s/it]
 13%|█▎        | 163/1250 [04:16<29:54,  1.65s/it]
 13%|█▎        | 164/1250 [04:17<30:57,  1.71s/it]
 13%|█▎        | 165/1250 [04:19<29:58,  1.66s/it]
 13%|█▎        | 166/1250 [04:20<28:18,  1.57s/it]
 13%|█▎        | 167/1250 [04:22<26:21,  1.46s/it]
 13%|█▎        | 168/1250 [04:23<23:44,  1.32s/it]
 14%|█▎        | 169/1250 [04:25<29:27,  1.63s/it]
 14%|█▎        | 170/1250 [04:27<30:05,  1.67s/it]
 14%|█▎        | 171/1250 [04:28<29:08,  1.62s/it]
 14%|█▍        | 172/1250 [04:30<27:20,  1.52s/it]
 14%|█▍        | 173/1250 [04:31<25:27,  1.42s/it]
 14%|█▍        | 174/1250 [04:32<22:57,  1.28s/it]
 14%|█▍        | 175/1250 [04:33<25:14,  1.41s/it]
 14%|█▍        | 176/1250 [04:36<32:49,  1.83s/it]
 14%|█▍        | 177/1250 [04:38<32:19,  1.81s/it]
 14%|█▍        | 178/1250 [04:39<30:54,  1.73s/it]
 14%|█▍        | 179/1250 [04:41<29:09,  1.63s/it]
 14%|█▍        | 180/1250 [04:42<26:40,  1.50s/it]
 14%|█▍        | 181/1250 [04:43<23:54,  1.34s/it]
 15%|█▍        | 182/1250 [04:45<28:58,  1.63s/it]
 15%|█▍        | 183/1250 [04:47<29:22,  1.65s/it]
 15%|█▍        | 184/1250 [04:49<28:39,  1.61s/it]
 15%|█▍        | 185/1250 [04:50<27:09,  1.53s/it]
 15%|█▍        | 186/1250 [04:51<25:10,  1.42s/it]
 15%|█▍        | 187/1250 [04:52<22:49,  1.29s/it]
 15%|█▌        | 188/1250 [04:55<29:02,  1.64s/it]
 15%|█▌        | 189/1250 [04:56<29:43,  1.68s/it]
 15%|█▌        | 190/1250 [04:58<28:55,  1.64s/it]
 15%|█▌        | 191/1250 [04:59<27:14,  1.54s/it]
 15%|█▌        | 192/1250 [05:00<25:07,  1.43s/it]
 15%|█▌        | 193/1250 [05:01<22:42,  1.29s/it]
 16%|█▌        | 194/1250 [05:04<27:52,  1.58s/it]
 16%|█▌        | 195/1250 [05:05<28:41,  1.63s/it]
 16%|█▌        | 196/1250 [05:07<28:06,  1.60s/it]
 16%|█▌        | 197/1250 [05:08<27:08,  1.55s/it]
 16%|█▌        | 198/1250 [05:09<25:40,  1.46s/it]
 16%|█▌        | 199/1250 [05:11<23:23,  1.34s/it]
 16%|█▌        | 200/1250 [05:12<26:23,  1.51s/it]
                                                  
 16%|█▌        | 200/1250 [05:12<26:23,  1.51s/it]
 16%|█▌        | 201/1250 [05:15<33:33,  1.92s/it]
 16%|█▌        | 202/1250 [05:17<32:28,  1.86s/it]
 16%|█▌        | 203/1250 [05:19<30:40,  1.76s/it]
 16%|█▋        | 204/1250 [05:20<28:51,  1.66s/it]
 16%|█▋        | 205/1250 [05:21<26:31,  1.52s/it]
 16%|█▋        | 206/1250 [05:22<23:42,  1.36s/it]
 17%|█▋        | 207/1250 [05:25<29:02,  1.67s/it]
 17%|█▋        | 208/1250 [05:26<29:31,  1.70s/it]
 17%|█▋        | 209/1250 [05:28<28:26,  1.64s/it]
 17%|█▋        | 210/1250 [05:29<26:54,  1.55s/it]
 17%|█▋        | 211/1250 [05:30<24:56,  1.44s/it]
 17%|█▋        | 212/1250 [05:31<22:33,  1.30s/it]
 17%|█▋        | 213/1250 [05:34<28:02,  1.62s/it]
 17%|█▋        | 214/1250 [05:35<28:39,  1.66s/it]
 17%|█▋        | 215/1250 [05:37<27:54,  1.62s/it]
 17%|█▋        | 216/1250 [05:38<26:40,  1.55s/it]
 17%|█▋        | 217/1250 [05:40<24:52,  1.44s/it]
 17%|█▋        | 218/1250 [05:41<22:36,  1.31s/it]
 18%|█▊        | 219/1250 [05:43<27:51,  1.62s/it]
 18%|█▊        | 220/1250 [05:45<28:23,  1.65s/it]
 18%|█▊        | 221/1250 [05:46<27:36,  1.61s/it]
 18%|█▊        | 222/1250 [05:47<26:09,  1.53s/it]
 18%|█▊        | 223/1250 [05:49<24:20,  1.42s/it]
 18%|█▊        | 224/1250 [05:50<22:20,  1.31s/it]
 18%|█▊        | 225/1250 [05:52<25:04,  1.47s/it]
 18%|█▊        | 226/1250 [05:54<32:16,  1.89s/it]
 18%|█▊        | 227/1250 [05:56<31:41,  1.86s/it]
 18%|█▊        | 228/1250 [05:58<30:00,  1.76s/it]
 18%|█▊        | 229/1250 [05:59<27:50,  1.64s/it]
 18%|█▊        | 230/1250 [06:00<25:26,  1.50s/it]
 18%|█▊        | 231/1250 [06:01<22:49,  1.34s/it]
 19%|█▊        | 232/1250 [06:04<27:42,  1.63s/it]
 19%|█▊        | 233/1250 [06:05<28:19,  1.67s/it]
 19%|█▊        | 234/1250 [06:07<27:47,  1.64s/it]
 19%|█▉        | 235/1250 [06:08<26:41,  1.58s/it]
 19%|█▉        | 236/1250 [06:10<25:01,  1.48s/it]
 19%|█▉        | 237/1250 [06:11<22:57,  1.36s/it]
 19%|█▉        | 238/1250 [06:13<28:17,  1.68s/it]
 19%|█▉        | 239/1250 [06:15<28:30,  1.69s/it]
 19%|█▉        | 240/1250 [06:16<27:31,  1.64s/it]
 19%|█▉        | 241/1250 [06:18<25:47,  1.53s/it]
 19%|█▉        | 242/1250 [06:19<23:55,  1.42s/it]
 19%|█▉        | 243/1250 [06:20<21:42,  1.29s/it]
 20%|█▉        | 244/1250 [06:22<27:29,  1.64s/it]
 20%|█▉        | 245/1250 [06:24<27:55,  1.67s/it]
 20%|█▉        | 246/1250 [06:25<27:02,  1.62s/it]
 20%|█▉        | 247/1250 [06:27<25:41,  1.54s/it]
 20%|█▉        | 248/1250 [06:28<23:48,  1.43s/it]
 20%|█▉        | 249/1250 [06:29<21:24,  1.28s/it]
 20%|██        | 250/1250 [06:31<23:19,  1.40s/it]
 20%|██        | 251/1250 [06:34<31:22,  1.88s/it]
 20%|██        | 252/1250 [06:35<30:47,  1.85s/it]
 20%|██        | 253/1250 [06:37<28:53,  1.74s/it]
 20%|██        | 254/1250 [06:38<26:57,  1.62s/it]
 20%|██        | 255/1250 [06:39<24:37,  1.48s/it]
 20%|██        | 256/1250 [06:40<22:05,  1.33s/it]
 21%|██        | 257/1250 [06:43<27:33,  1.66s/it]
 21%|██        | 258/1250 [06:45<28:01,  1.69s/it]
 21%|██        | 259/1250 [06:46<27:31,  1.67s/it]
 21%|██        | 260/1250 [06:48<26:14,  1.59s/it]
 21%|██        | 261/1250 [06:49<24:25,  1.48s/it]
 21%|██        | 262/1250 [06:50<21:49,  1.33s/it]
 21%|██        | 263/1250 [06:52<26:46,  1.63s/it]
 21%|██        | 264/1250 [06:54<27:14,  1.66s/it]
 21%|██        | 265/1250 [06:55<26:29,  1.61s/it]
 21%|██▏       | 266/1250 [06:57<25:03,  1.53s/it]
 21%|██▏       | 267/1250 [06:58<23:19,  1.42s/it]
 21%|██▏       | 268/1250 [06:59<21:12,  1.30s/it]
 22%|██▏       | 269/1250 [07:01<25:45,  1.58s/it]
 22%|██▏       | 270/1250 [07:03<26:41,  1.63s/it]
 22%|██▏       | 271/1250 [07:04<26:09,  1.60s/it]
 22%|██▏       | 272/1250 [07:06<24:53,  1.53s/it]
 22%|██▏       | 273/1250 [07:07<23:15,  1.43s/it]
 22%|██▏       | 274/1250 [07:08<21:06,  1.30s/it]
 22%|██▏       | 275/1250 [07:10<22:48,  1.40s/it]
 22%|██▏       | 276/1250 [07:12<29:24,  1.81s/it]
 22%|██▏       | 277/1250 [07:14<29:19,  1.81s/it]
 22%|██▏       | 278/1250 [07:16<27:58,  1.73s/it]
 22%|██▏       | 279/1250 [07:17<26:26,  1.63s/it]
 22%|██▏       | 280/1250 [07:18<24:16,  1.50s/it]
 22%|██▏       | 281/1250 [07:19<21:57,  1.36s/it]
 23%|██▎       | 282/1250 [07:22<26:51,  1.66s/it]
 23%|██▎       | 283/1250 [07:23<27:18,  1.69s/it]
 23%|██▎       | 284/1250 [07:25<26:33,  1.65s/it]
 23%|██▎       | 285/1250 [07:26<25:13,  1.57s/it]
 23%|██▎       | 286/1250 [07:28<23:27,  1.46s/it]
 23%|██▎       | 287/1250 [07:29<21:21,  1.33s/it]
 23%|██▎       | 288/1250 [07:31<26:40,  1.66s/it]
 23%|██▎       | 289/1250 [07:33<26:49,  1.68s/it]
 23%|██▎       | 290/1250 [07:34<26:15,  1.64s/it]
 23%|██▎       | 291/1250 [07:36<24:52,  1.56s/it]
 23%|██▎       | 292/1250 [07:37<22:56,  1.44s/it]
 23%|██▎       | 293/1250 [07:38<20:36,  1.29s/it]
 24%|██▎       | 294/1250 [07:40<25:20,  1.59s/it]
 24%|██▎       | 295/1250 [07:42<25:45,  1.62s/it]
 24%|██▎       | 296/1250 [07:43<24:59,  1.57s/it]
 24%|██▍       | 297/1250 [07:44<23:42,  1.49s/it]
 24%|██▍       | 298/1250 [07:46<22:08,  1.40s/it]
 24%|██▍       | 299/1250 [07:47<20:04,  1.27s/it]
 24%|██▍       | 300/1250 [07:48<22:58,  1.45s/it]
                                                  
 24%|██▍       | 300/1250 [07:48<22:58,  1.45s/it]
 24%|██▍       | 301/1250 [07:51<29:17,  1.85s/it]
 24%|██▍       | 302/1250 [07:53<28:36,  1.81s/it]
 24%|██▍       | 303/1250 [07:55<27:16,  1.73s/it]
 24%|██▍       | 304/1250 [07:56<25:46,  1.63s/it]
 24%|██▍       | 305/1250 [07:57<23:38,  1.50s/it]
 24%|██▍       | 306/1250 [07:58<21:16,  1.35s/it]
 25%|██▍       | 307/1250 [08:00<25:45,  1.64s/it]
 25%|██▍       | 308/1250 [08:02<26:24,  1.68s/it]
 25%|██▍       | 309/1250 [08:04<25:52,  1.65s/it]
 25%|██▍       | 310/1250 [08:05<24:30,  1.56s/it]
 25%|██▍       | 311/1250 [08:06<22:52,  1.46s/it]
 25%|██▍       | 312/1250 [08:07<20:47,  1.33s/it]
 25%|██▌       | 313/1250 [08:10<26:16,  1.68s/it]
 25%|██▌       | 314/1250 [08:12<26:45,  1.72s/it]
 25%|██▌       | 315/1250 [08:13<25:53,  1.66s/it]
 25%|██▌       | 316/1250 [08:15<24:24,  1.57s/it]
 25%|██▌       | 317/1250 [08:16<22:36,  1.45s/it]
 25%|█��▌       | 318/1250 [08:17<20:24,  1.31s/it]
 26%|██▌       | 319/1250 [08:19<25:10,  1.62s/it]
 26%|██▌       | 320/1250 [08:21<25:30,  1.65s/it]
 26%|██▌       | 321/1250 [08:22<24:42,  1.60s/it]
 26%|██▌       | 322/1250 [08:24<23:36,  1.53s/it]
 26%|██▌       | 323/1250 [08:25<21:59,  1.42s/it]
 26%|██▌       | 324/1250 [08:26<19:52,  1.29s/it]
 26%|██▌       | 325/1250 [08:28<22:51,  1.48s/it]
 26%|██▌       | 326/1250 [08:31<28:59,  1.88s/it]
 26%|██▌       | 327/1250 [08:32<28:15,  1.84s/it]
 26%|██▌       | 328/1250 [08:34<26:38,  1.73s/it]
 26%|██▋       | 329/1250 [08:35<24:51,  1.62s/it]
 26%|██▋       | 330/1250 [08:36<22:40,  1.48s/it]
 26%|██▋       | 331/1250 [08:37<20:21,  1.33s/it]
 27%|██▋       | 332/1250 [08:40<24:40,  1.61s/it]
 27%|██▋       | 333/1250 [08:41<25:06,  1.64s/it]
 27%|██▋       | 334/1250 [08:43<24:38,  1.61s/it]
 27%|██▋       | 335/1250 [08:44<23:33,  1.54s/it]
 27%|██▋       | 336/1250 [08:45<21:48,  1.43s/it]
 27%|██▋       | 337/1250 [08:46<19:37,  1.29s/it]
 27%|██▋       | 338/1250 [08:49<24:31,  1.61s/it]
 27%|██▋       | 339/1250 [08:50<25:13,  1.66s/it]
 27%|██▋       | 340/1250 [08:52<24:29,  1.61s/it]
 27%|██▋       | 341/1250 [08:53<23:28,  1.55s/it]
 27%|██▋       | 342/1250 [08:55<21:50,  1.44s/it]
 27%|██▋       | 343/1250 [08:56<20:01,  1.32s/it]
 28%|██▊       | 344/1250 [08:58<24:53,  1.65s/it]
 28%|██▊       | 345/1250 [09:00<25:11,  1.67s/it]
 28%|██▊       | 346/1250 [09:01<24:32,  1.63s/it]
 28%|██▊       | 347/1250 [09:03<23:18,  1.55s/it]
 28%|██▊       | 348/1250 [09:04<21:27,  1.43s/it]
 28%|██▊       | 349/1250 [09:05<19:20,  1.29s/it]
 28%|██▊       | 350/1250 [09:07<22:39,  1.51s/it]
 28%|██▊       | 351/1250 [09:10<29:06,  1.94s/it]
 28%|██▊       | 352/1250 [09:12<28:24,  1.90s/it]
 28%|██▊       | 353/1250 [09:13<26:48,  1.79s/it]
 28%|██▊       | 354/1250 [09:14<24:54,  1.67s/it]
 28%|██▊       | 355/1250 [09:16<22:40,  1.52s/it]
 28%|██▊       | 356/1250 [09:17<20:29,  1.38s/it]
 29%|██▊       | 357/1250 [09:19<24:22,  1.64s/it]
 29%|██▊       | 358/1250 [09:21<24:35,  1.65s/it]
 29%|██▊       | 359/1250 [09:22<23:54,  1.61s/it]
 29%|██▉       | 360/1250 [09:23<22:44,  1.53s/it]
 29%|██▉       | 361/1250 [09:25<21:14,  1.43s/it]
 29%|██▉       | 362/1250 [09:26<19:13,  1.30s/it]
 29%|██▉       | 363/1250 [09:28<23:56,  1.62s/it]
 29%|██▉       | 364/1250 [09:30<24:15,  1.64s/it]
 29%|██▉       | 365/1250 [09:31<23:39,  1.60s/it]
 29%|██▉       | 366/1250 [09:33<22:30,  1.53s/it]
 29%|██▉       | 367/1250 [09:34<21:11,  1.44s/it]
 29%|██▉       | 368/1250 [09:35<19:05,  1.30s/it]
 30%|██▉       | 369/1250 [09:37<24:00,  1.63s/it]
 30%|██▉       | 370/1250 [09:39<24:30,  1.67s/it]
 30%|██▉       | 371/1250 [09:40<23:53,  1.63s/it]
 30%|██▉       | 372/1250 [09:42<22:41,  1.55s/it]
 30%|██▉       | 373/1250 [09:43<20:58,  1.44s/it]
 30%|██▉       | 374/1250 [09:44<18:54,  1.30s/it]
 30%|███       | 375/1250 [09:46<20:17,  1.39s/it]
 30%|███       | 376/1250 [09:48<26:28,  1.82s/it]
 30%|███       | 377/1250 [09:50<25:49,  1.78s/it]
 30%|███       | 378/1250 [09:52<24:25,  1.68s/it]
 30%|███       | 379/1250 [09:53<22:51,  1.57s/it]
 30%|███       | 380/1250 [09:54<21:04,  1.45s/it]
 30%|███       | 381/1250 [09:55<18:55,  1.31s/it]
 31%|███       | 382/1250 [09:57<23:55,  1.65s/it]
 31%|███       | 383/1250 [09:59<24:26,  1.69s/it]
 31%|███       | 384/1250 [10:01<23:44,  1.64s/it]
 31%|███       | 385/1250 [10:02<22:36,  1.57s/it]
 31%|███       | 386/1250 [10:03<21:00,  1.46s/it]
 31%|███       | 387/1250 [10:04<19:05,  1.33s/it]
 31%|███       | 388/1250 [10:07<22:40,  1.58s/it]
 31%|███       | 389/1250 [10:08<23:05,  1.61s/it]
 31%|███       | 390/1250 [10:10<22:56,  1.60s/it]
 31%|███▏      | 391/1250 [10:11<22:10,  1.55s/it]
 31%|███▏      | 392/1250 [10:12<20:33,  1.44s/it]
 31%|███▏      | 393/1250 [10:13<18:37,  1.30s/it]
 32%|███▏      | 394/1250 [10:16<23:25,  1.64s/it]
 32%|███▏      | 395/1250 [10:18<23:45,  1.67s/it]
 32%|███▏      | 396/1250 [10:19<23:08,  1.63s/it]
 32%|███▏      | 397/1250 [10:20<21:55,  1.54s/it]
 32%|███▏      | 398/1250 [10:22<20:16,  1.43s/it]
 32%|███▏      | 399/1250 [10:23<18:21,  1.29s/it]
 32%|███▏      | 400/1250 [10:24<20:34,  1.45s/it]
                                                  
 32%|███▏      | 400/1250 [10:24<20:34,  1.45s/it]
 32%|███▏      | 401/1250 [10:27<26:02,  1.84s/it]
 32%|███▏      | 402/1250 [10:29<25:43,  1.82s/it]
 32%|███▏      | 403/1250 [10:31<24:36,  1.74s/it]
 32%|███▏      | 404/1250 [10:32<22:58,  1.63s/it]
 32%|███▏      | 405/1250 [10:33<21:12,  1.51s/it]
 32%|███▏      | 406/1250 [10:34<19:08,  1.36s/it]
 33%|███▎      | 407/1250 [10:37<23:55,  1.70s/it]
 33%|███▎      | 408/1250 [10:38<24:11,  1.72s/it]
 33%|███▎      | 409/1250 [10:40<23:27,  1.67s/it]
 33%|███▎      | 410/1250 [10:41<22:04,  1.58s/it]
 33%|███▎      | 411/1250 [10:42<20:25,  1.46s/it]
 33%|███▎      | 412/1250 [10:43<18:26,  1.32s/it]
 33%|███▎      | 413/1250 [10:46<22:59,  1.65s/it]
 33%|███▎      | 414/1250 [10:48<23:20,  1.68s/it]
 33%|███▎      | 415/1250 [10:49<22:33,  1.62s/it]
 33%|███▎      | 416/1250 [10:50<21:15,  1.53s/it]
 33%|███▎      | 417/1250 [10:52<19:25,  1.40s/it]
 33%|███▎      | 418/1250 [10:52<17:26,  1.26s/it]
 34%|███▎      | 419/1250 [10:55<21:48,  1.57s/it]
 34%|███▎      | 420/1250 [10:56<22:18,  1.61s/it]
 34%|███▎      | 421/1250 [10:58<21:51,  1.58s/it]
 34%|███▍      | 422/1250 [10:59<21:12,  1.54s/it]
 34%|███▍      | 423/1250 [11:01<19:59,  1.45s/it]
 34%|███▍      | 424/1250 [11:02<18:08,  1.32s/it]
 34%|███▍      | 425/1250 [11:03<19:17,  1.40s/it]
 34%|███▍      | 426/1250 [11:06<25:20,  1.85s/it]
 34%|███▍      | 427/1250 [11:08<24:54,  1.82s/it]
 34%|███▍      | 428/1250 [11:09<23:45,  1.73s/it]
 34%|███▍      | 429/1250 [11:11<22:22,  1.64s/it]
 34%|███▍      | 430/1250 [11:12<20:29,  1.50s/it]
 34%|███▍      | 431/1250 [11:13<18:27,  1.35s/it]
 35%|███▍      | 432/1250 [11:15<21:54,  1.61s/it]
 35%|███▍      | 433/1250 [11:17<22:24,  1.65s/it]
 35%|███▍      | 434/1250 [11:18<21:43,  1.60s/it]
 35%|███▍      | 435/1250 [11:20<20:37,  1.52s/it]
 35%|███▍      | 436/1250 [11:21<19:27,  1.43s/it]
 35%|███▍      | 437/1250 [11:22<17:33,  1.30s/it]
 35%|███▌      | 438/1250 [11:24<21:45,  1.61s/it]
 35%|███▌      | 439/1250 [11:26<22:22,  1.65s/it]
 35%|███▌      | 440/1250 [11:28<21:56,  1.62s/it]
 35%|███▌      | 441/1250 [11:29<21:01,  1.56s/it]
 35%|███▌      | 442/1250 [11:30<19:33,  1.45s/it]
 35%|███▌      | 443/1250 [11:31<17:42,  1.32s/it]
 36%|███▌      | 444/1250 [11:34<22:11,  1.65s/it]
 36%|███▌      | 445/1250 [11:35<22:31,  1.68s/it]
 36%|███▌      | 446/1250 [11:37<22:06,  1.65s/it]
 36%|███▌      | 447/1250 [11:38<20:52,  1.56s/it]
 36%|███▌      | 448/1250 [11:40<19:07,  1.43s/it]
 36%|███▌      | 449/1250 [11:40<17:11,  1.29s/it]
 36%|███▌      | 450/1250 [11:42<18:49,  1.41s/it]
 36%|███▌      | 451/1250 [11:45<24:40,  1.85s/it]
 36%|███▌      | 452/1250 [11:47<24:24,  1.84s/it]
 36%|███▌      | 453/1250 [11:48<23:15,  1.75s/it]
 36%|███▋      | 454/1250 [11:50<21:50,  1.65s/it]
 36%|███▋      | 455/1250 [11:51<20:05,  1.52s/it]
 36%|███▋      | 456/1250 [11:52<17:55,  1.35s/it]
 37%|███▋      | 457/1250 [11:54<22:22,  1.69s/it]
 37%|███▋      | 458/1250 [11:56<22:42,  1.72s/it]
 37%|███▋      | 459/1250 [11:58<21:46,  1.65s/it]
 37%|███▋      | 460/1250 [11:59<20:25,  1.55s/it]
 37%|███▋      | 461/1250 [12:00<18:57,  1.44s/it]
 37%|███▋      | 462/1250 [12:01<17:05,  1.30s/it]
 37%|███▋      | 463/1250 [12:03<20:37,  1.57s/it]
 37%|███▋      | 464/1250 [12:05<21:05,  1.61s/it]
 37%|███▋      | 465/1250 [12:07<20:35,  1.57s/it]
 37%|███▋      | 466/1250 [12:08<19:39,  1.50s/it]
 37%|███▋      | 467/1250 [12:09<18:22,  1.41s/it]
 37%|███▋      | 468/1250 [12:10<16:57,  1.30s/it]
 38%|███▊      | 469/1250 [12:12<20:46,  1.60s/it]
 38%|███▊      | 470/1250 [12:14<21:07,  1.62s/it]
 38%|███▊      | 471/1250 [12:16<20:47,  1.60s/it]
 38%|███▊      | 472/1250 [12:17<19:43,  1.52s/it]
 38%|███▊      | 473/1250 [12:18<18:18,  1.41s/it]
 38%|███▊      | 474/1250 [12:19<16:37,  1.29s/it]
 38%|███▊      | 475/1250 [12:21<18:49,  1.46s/it]
 38%|███▊      | 476/1250 [12:24<24:27,  1.90s/it]
 38%|███▊      | 477/1250 [12:26<24:00,  1.86s/it]
 38%|███▊      | 478/1250 [12:27<22:39,  1.76s/it]
 38%|███▊      | 479/1250 [12:29<21:01,  1.64s/it]
 38%|███▊      | 480/1250 [12:30<18:58,  1.48s/it]
 38%|███▊      | 481/1250 [12:31<16:44,  1.31s/it]
 39%|███▊      | 482/1250 [12:33<20:30,  1.60s/it]
 39%|███▊      | 483/1250 [12:35<20:39,  1.62s/it]
 39%|███▊      | 484/1250 [12:36<20:30,  1.61s/it]
 39%|███▉      | 485/1250 [12:38<19:36,  1.54s/it]
 39%|███▉      | 486/1250 [12:39<18:15,  1.43s/it]
 39%|███▉      | 487/1250 [12:40<16:37,  1.31s/it]
 39%|███▉      | 488/1250 [12:42<20:55,  1.65s/it]
 39%|███▉      | 489/1250 [12:44<21:17,  1.68s/it]
 39%|███▉      | 490/1250 [12:45<20:34,  1.62s/it]
 39%|███▉      | 491/1250 [12:47<19:28,  1.54s/it]
 39%|███▉      | 492/1250 [12:48<18:28,  1.46s/it]
 39%|███▉      | 493/1250 [12:49<16:39,  1.32s/it]
 40%|███▉      | 494/1250 [12:51<20:23,  1.62s/it]
 40%|███▉      | 495/1250 [12:53<20:39,  1.64s/it]
 40%|███▉      | 496/1250 [12:55<20:11,  1.61s/it]
 40%|███▉      | 497/1250 [12:56<19:09,  1.53s/it]
 40%|███▉      | 498/1250 [12:57<17:44,  1.42s/it]
 40%|███▉      | 499/1250 [12:58<15:57,  1.28s/it]
 40%|████      | 500/1250 [13:00<17:17,  1.38s/it]
                                                  
 40%|████      | 500/1250 [13:00<17:17,  1.38s/it]The following columns in the evaluation set  don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.
+***** Running Evaluation *****
+  Num examples = 341
+  Batch size = 8
+{'loss': 20.0794, 'learning_rate': 3.675e-06, 'epoch': 3.98}
+{'loss': 10.5776, 'learning_rate': 7.425e-06, 'epoch': 7.98}
+{'loss': 6.6033, 'learning_rate': 1.1174999999999999e-05, 'epoch': 11.98}
+{'loss': 5.3857, 'learning_rate': 1.4925e-05, 'epoch': 15.98}
+{'loss': 4.4431, 'learning_rate': 1.8675e-05, 'epoch': 19.98}
+
+
  0%|          | 0/43 [00:00<?, ?it/s]
+
  5%|▍         | 2/43 [00:00<00:04,  9.99it/s]
+
  7%|▋         | 3/43 [00:00<00:07,  5.48it/s]
+
  9%|▉         | 4/43 [00:00<00:09,  4.26it/s]
+
 12%|█▏        | 5/43 [00:01<00:09,  4.09it/s]
+
 14%|█▍        | 6/43 [00:01<00:09,  3.90it/s]
+
 16%|█▋        | 7/43 [00:01<00:09,  3.85it/s]
+
 19%|█▊        | 8/43 [00:01<00:09,  3.77it/s]
+
 21%|██        | 9/43 [00:02<00:09,  3.47it/s]
+
 23%|██▎       | 10/43 [00:02<00:08,  3.74it/s]
+
 26%|██▌       | 11/43 [00:02<00:09,  3.54it/s]
+
 28%|██▊       | 12/43 [00:03<00:08,  3.77it/s]
+
 30%|███       | 13/43 [00:03<00:07,  3.86it/s]
+
 33%|███▎      | 14/43 [00:03<00:08,  3.48it/s]
+
 35%|███▍      | 15/43 [00:04<00:09,  2.99it/s]
+
 37%|███▋      | 16/43 [00:04<00:08,  3.05it/s]
+
 40%|███▉      | 17/43 [00:04<00:08,  3.14it/s]
+
 42%|████▏     | 18/43 [00:05<00:08,  3.05it/s]
+
 44%|████▍     | 19/43 [00:05<00:08,  2.85it/s]
+
 47%|████▋     | 20/43 [00:05<00:09,  2.54it/s]
+
 49%|████▉     | 21/43 [00:06<00:08,  2.66it/s]
+
 51%|█████     | 22/43 [00:06<00:07,  2.70it/s]
+
 53%|█████▎    | 23/43 [00:06<00:07,  2.82it/s]
+
 56%|█████▌    | 24/43 [00:07<00:06,  3.09it/s]
+
 58%|█████▊    | 25/43 [00:07<00:05,  3.22it/s]
+
 60%|██████    | 26/43 [00:07<00:05,  3.21it/s]
+
 63%|██████▎   | 27/43 [00:07<00:04,  3.59it/s]
+
 65%|██████▌   | 28/43 [00:08<00:04,  3.65it/s]
+
 67%|██████▋   | 29/43 [00:08<00:03,  3.76it/s]
+
 70%|██████▉   | 30/43 [00:08<00:03,  3.83it/s]
+
 72%|███████▏  | 31/43 [00:08<00:03,  3.79it/s]
+
 74%|███████▍  | 32/43 [00:09<00:02,  3.85it/s]
+
 77%|███████▋  | 33/43 [00:09<00:02,  3.54it/s]
+
 79%|███████▉  | 34/43 [00:09<00:02,  3.51it/s]
+
 81%|████████▏ | 35/43 [00:10<00:02,  3.51it/s]
+
 84%|████████▎ | 36/43 [00:10<00:02,  3.42it/s]
+
 86%|████████▌ | 37/43 [00:10<00:01,  3.56it/s]
+
 88%|████████▊ | 38/43 [00:11<00:01,  3.32it/s]
+
 91%|█████████ | 39/43 [00:11<00:01,  3.38it/s]
+
 93%|█████████▎| 40/43 [00:11<00:00,  3.27it/s]
+
 95%|█████████▌| 41/43 [00:11<00:00,  3.37it/s]
+
 98%|█████████▊| 42/43 [00:12<00:00,  3.19it/s]
+
                                                  
+
                                               
 40%|████      | 500/1250 [13:13<17:17,  1.38s/it]
+
+
                                               Saving model checkpoint to ./checkpoint-500
+Configuration saved in ./checkpoint-500/config.json
+Model weights saved in ./checkpoint-500/pytorch_model.bin
+Configuration saved in ./checkpoint-500/preprocessor_config.json
+Configuration saved in ./preprocessor_config.json

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 sha256:4b3c7996c48bbce91a11a7b8efbb9dbd48a52cc46b992569ecdef9e54e1180da
+size 1262173745

run.sh

@@ -0,0 +1,35 @@
+python run_speech_recognition_ctc.py \
+--dataset_name="mozilla-foundation/common_voice_8_0" \
+--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
+--dataset_config_name="ur" \
+--output_dir="./" \
+--overwrite_output_dir \
+--num_train_epochs="50" \
+--per_device_train_batch_size="8" \
+--per_device_eval_batch_size="8" \
+--gradient_accumulation_steps="4" \
+--learning_rate="7.5e-5" \
+--warmup_steps="2000" \
+--length_column_name="input_length" \
+--evaluation_strategy="steps" \
+--text_column_name="sentence" \
+--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
+--save_steps="500" \
+--eval_steps="500" \
+--logging_steps="100" \
+--layerdrop="0.0" \
+--activation_dropout="0.1" \
+--save_total_limit="3" \
+--freeze_feature_encoder \
+--feat_proj_dropout="0.0" \
+--mask_time_prob="0.75" \
+--mask_time_length="10" \
+--mask_feature_prob="0.25" \
+--mask_feature_length="64" \
+--gradient_checkpointing \
+--use_auth_token \
+--fp16 \
+--group_by_length \
+--do_train=1 \
+--do_eval=1 \
+--push_to_hub

run_speech_recognition_ctc.py

@@ -0,0 +1,737 @@
+#!/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("4.17.0.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+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:`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 = (
+        f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
+    )
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                os.remove(vocab_file)
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                vocab_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "feat_proj_dropout": model_args.feat_proj_dropout,
+            "attention_dropout": model_args.attention_dropout,
+            "hidden_dropout": model_args.hidden_dropout,
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "activation_dropout": model_args.activation_dropout,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+    )
+
+    # 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 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]", "additional_special_tokens": [{"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}]}

tokenizer_config.json

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

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ac9126d8e40ef2b4777d784e17705def6720cee411288fde206b31f5e389f9a
+size 2991

vocab.json

@@ -0,0 +1 @@
+{"،": 1, "؟": 2, "ء": 3, "آ": 4, "ؤ": 5, "ئ": 6, "ا": 7, "ب": 8, "ت": 9, "ث": 10, "ج": 11, "ح": 12, "خ": 13, "د": 14, "ذ": 15, "ر": 16, "ز": 17, "س": 18, "ش": 19, "ص": 20, "ض": 21, "ط": 22, "ظ": 23, "ع": 24, "غ": 25, "ف": 26, "ق": 27, "ل": 28, "م": 29, "ن": 30, "و": 31, "ى": 32, "ي": 33, "ً": 34, "َ": 35, "ُ": 36, "ِ": 37, "ّ": 38, "ٓ": 39, "ٔ": 40, "ٰ": 41, "ٹ": 42, "پ": 43, "چ": 44, "ڈ": 45, "ڑ": 46, "ژ": 47, "ک": 48, "گ": 49, "ں": 50, "ھ": 51, "ہ": 52, "ۂ": 53, "ی": 54, "ے": 55, "۔": 56, "|": 0, "[UNK]": 57, "[PAD]": 58}