Training in progress, step 500

.gitignore

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

added_tokens.json

@@ -0,0 +1 @@
+{"<s>": 1205, "</s>": 1206}

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": 1204,
+  "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": 1207,
+  "xvector_output_dim": 512
+}

nohup.out

@@ -0,0 +1,569 @@
+01/31/2022 07:15:59 - WARNING - __main__ - Process rank: -1, device: cuda:0, n_gpu: 1distributed training: False, 16-bits training: True
+01/31/2022 07:15:59 - 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/Jan31_07-15-59_job-2c68f48a-2d5d-4013-9043-3f2cb25f3ff6,
+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=['tensorboard'],
+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,
+)
+01/31/2022 07:16:01 - WARNING - datasets.builder - Reusing dataset zeroth_korean_asr (/workspace/.cache/huggingface/datasets/kresnik___zeroth_korean_asr/clean/1.0.1/f6cf96a53d5512525e3113bab8048d36ce268658d6e0c40d45f65dfa3f0bc343)
+01/31/2022 07:16:03 - WARNING - datasets.builder - Reusing dataset zeroth_korean_asr (/workspace/.cache/huggingface/datasets/kresnik___zeroth_korean_asr/clean/1.0.1/f6cf96a53d5512525e3113bab8048d36ce268658d6e0c40d45f65dfa3f0bc343)
+
+
+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_config.json. We won't load it.
+Didn't find file ./added_tokens.json. We won't load it.
+Didn't find file ./special_tokens_map.json. We won't load it.
+Didn't find file ./tokenizer.json. We won't load it.
+loading file ./vocab.json
+loading file None
+loading file None
+loading file None
+loading file None
+file ./config.json not found
+Adding <s> to the vocabulary
+Adding </s> to the vocabulary
+Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
+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.bias', 'project_q.bias', 'quantizer.weight_proj.weight', 'project_hid.bias', 'project_q.weight', 'quantizer.codevectors', '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.bias', 'lm_head.weight']
+You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
+
+
+
  0%|          | 0/23 [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": 1204,
+  "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": 1207,
+  "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/wav2vec2-xls-r-300m-korean/./ is already a clone of https://huggingface.co/w11wo/wav2vec2-xls-r-300m-korean. Make sure you pull the latest changes with `repo.git_pull()`.
+01/31/2022 07:18:18 - WARNING - huggingface_hub.repository - /workspace/wav2vec2-xls-r-300m-korean/./ is already a clone of https://huggingface.co/w11wo/wav2vec2-xls-r-300m-korean. 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 = 22262
+  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 = 34750
+
  0%|          | 0/34750 [00:00<?, ?it/s]
  0%|          | 1/34750 [00:06<64:28:32,  6.68s/it]
  0%|          | 2/34750 [00:12<57:43:53,  5.98s/it]
  0%|          | 3/34750 [00:17<54:31:03,  5.65s/it]
  0%|          | 4/34750 [00:22<50:51:50,  5.27s/it]
  0%|          | 5/34750 [00:26<48:33:10,  5.03s/it]
  0%|          | 6/34750 [00:31<46:50:38,  4.85s/it]
  0%|          | 7/34750 [00:35<45:36:23,  4.73s/it]
  0%|          | 8/34750 [00:40<44:29:00,  4.61s/it]
  0%|          | 9/34750 [00:44<43:10:17,  4.47s/it]
  0%|          | 10/34750 [00:48<42:06:37,  4.36s/it]
  0%|          | 11/34750 [00:52<40:57:19,  4.24s/it]
  0%|          | 12/34750 [00:56<39:38:33,  4.11s/it]
  0%|          | 13/34750 [00:59<38:31:48,  3.99s/it]
  0%|          | 14/34750 [01:03<37:54:45,  3.93s/it]
  0%|          | 15/34750 [01:07<36:35:46,  3.79s/it]
  0%|          | 16/34750 [01:10<35:45:25,  3.71s/it]
  0%|          | 17/34750 [01:13<34:45:16,  3.60s/it]
  0%|          | 18/34750 [01:17<33:59:39,  3.52s/it]
  0%|          | 19/34750 [01:20<33:05:56,  3.43s/it]
  0%|          | 20/34750 [01:23<32:33:40,  3.38s/it]
  0%|          | 21/34750 [01:26<32:01:32,  3.32s/it]
  0%|          | 22/34750 [01:30<31:21:02,  3.25s/it]
  0%|          | 23/34750 [01:33<30:52:11,  3.20s/it]
  0%|          | 24/34750 [01:36<30:24:22,  3.15s/it]
  0%|          | 25/34750 [01:39<29:53:04,  3.10s/it]
  0%|          | 26/34750 [01:42<29:25:40,  3.05s/it]
  0%|          | 27/34750 [01:44<28:34:56,  2.96s/it]
  0%|          | 28/34750 [01:47<28:11:40,  2.92s/it]
  0%|          | 29/34750 [01:50<27:54:42,  2.89s/it]
  0%|          | 30/34750 [01:53<27:27:22,  2.85s/it]
  0%|          | 31/34750 [01:55<27:01:28,  2.80s/it]
  0%|          | 32/34750 [01:58<26:40:58,  2.77s/it]
  0%|          | 33/34750 [02:01<26:24:14,  2.74s/it]
  0%|          | 34/34750 [02:03<25:52:04,  2.68s/it]
  0%|          | 35/34750 [02:06<25:43:25,  2.67s/it]
  0%|          | 36/34750 [02:08<25:11:08,  2.61s/it]
  0%|          | 37/34750 [02:11<24:28:28,  2.54s/it]
  0%|          | 38/34750 [02:13<23:51:25,  2.47s/it]
  0%|          | 39/34750 [02:15<23:19:33,  2.42s/it]
  0%|          | 40/34750 [02:18<22:52:11,  2.37s/it]
  0%|          | 41/34750 [02:20<22:39:10,  2.35s/it]
  0%|          | 42/34750 [02:22<22:09:25,  2.30s/it]
  0%|          | 43/34750 [02:24<21:47:00,  2.26s/it]
  0%|          | 44/34750 [02:26<21:18:01,  2.21s/it]
  0%|          | 45/34750 [02:28<20:48:32,  2.16s/it]
  0%|          | 46/34750 [02:31<20:29:08,  2.13s/it]
  0%|          | 47/34750 [02:32<20:01:46,  2.08s/it]
  0%|          | 48/34750 [02:34<19:24:47,  2.01s/it]
  0%|          | 49/34750 [02:36<18:42:30,  1.94s/it]
  0%|          | 50/34750 [02:38<17:57:24,  1.86s/it]
  0%|          | 51/34750 [02:44<31:34:21,  3.28s/it]
  0%|          | 52/34750 [02:50<38:06:08,  3.95s/it]
  0%|          | 53/34750 [02:55<41:11:56,  4.27s/it]
  0%|          | 54/34750 [03:00<42:58:47,  4.46s/it]
  0%|          | 55/34750 [03:04<43:32:12,  4.52s/it]
  0%|          | 56/34750 [03:09<43:38:21,  4.53s/it]
  0%|          | 57/34750 [03:13<43:18:06,  4.49s/it]
  0%|          | 58/34750 [03:18<42:30:15,  4.41s/it]
  0%|          | 59/34750 [03:22<42:02:18,  4.36s/it]
  0%|          | 60/34750 [03:26<41:31:48,  4.31s/it]
  0%|          | 61/34750 [03:30<40:32:29,  4.21s/it]
  0%|          | 62/34750 [03:34<39:22:08,  4.09s/it]
  0%|          | 63/34750 [03:38<38:18:34,  3.98s/it]
  0%|          | 64/34750 [03:41<37:37:25,  3.90s/it]
  0%|          | 65/34750 [03:45<36:45:09,  3.81s/it]
  0%|          | 66/34750 [03:48<35:44:42,  3.71s/it]
  0%|          | 67/34750 [03:52<35:21:38,  3.67s/it]
  0%|          | 68/34750 [03:55<34:38:18,  3.60s/it]
  0%|          | 69/34750 [03:59<33:35:03,  3.49s/it]
  0%|          | 70/34750 [04:02<32:57:42,  3.42s/it]
  0%|          | 71/34750 [04:05<32:30:58,  3.38s/it]
  0%|          | 72/34750 [04:08<31:58:21,  3.32s/it]
  0%|          | 73/34750 [04:12<31:39:22,  3.29s/it]
  0%|          | 74/34750 [04:15<30:59:41,  3.22s/it]
  0%|          | 75/34750 [04:18<30:22:31,  3.15s/it]
  0%|          | 76/34750 [04:21<30:16:16,  3.14s/it]
  0%|          | 77/34750 [04:24<30:00:28,  3.12s/it]
  0%|          | 78/34750 [04:27<29:35:28,  3.07s/it]
  0%|          | 79/34750 [04:30<28:52:03,  3.00s/it]
  0%|          | 80/34750 [04:32<28:15:57,  2.94s/it]
  0%|          | 81/34750 [04:35<27:39:57,  2.87s/it]
  0%|          | 82/34750 [04:38<27:02:17,  2.81s/it]
  0%|          | 83/34750 [04:40<26:42:09,  2.77s/it]
  0%|          | 84/34750 [04:43<26:15:39,  2.73s/it]
  0%|          | 85/34750 [04:46<26:03:43,  2.71s/it]
  0%|          | 86/34750 [04:48<25:34:47,  2.66s/it]
  0%|          | 87/34750 [04:51<25:14:44,  2.62s/it]
  0%|          | 88/34750 [04:53<24:52:15,  2.58s/it]
  0%|          | 89/34750 [04:56<24:13:33,  2.52s/it]
  0%|          | 90/34750 [04:58<23:45:50,  2.47s/it]
  0%|          | 91/34750 [05:00<23:15:35,  2.42s/it]
  0%|          | 92/34750 [05:03<22:43:29,  2.36s/it]
  0%|          | 93/34750 [05:05<22:28:51,  2.34s/it]
  0%|          | 94/34750 [05:07<21:54:02,  2.28s/it]
  0%|          | 95/34750 [05:09<21:18:57,  2.21s/it]
  0%|          | 96/34750 [05:11<20:44:18,  2.15s/it]
  0%|          | 97/34750 [05:13<20:04:09,  2.08s/it]
  0%|          | 98/34750 [05:15<19:41:42,  2.05s/it]
  0%|          | 99/34750 [05:17<19:05:17,  1.98s/it]
  0%|          | 100/34750 [05:18<18:21:38,  1.91s/it]
                                                      
  0%|          | 100/34750 [05:18<18:21:38,  1.91s/it]
  0%|          | 101/34750 [05:25<31:59:06,  3.32s/it]
  0%|          | 102/34750 [05:31<38:44:13,  4.02s/it]
  0%|          | 103/34750 [05:36<42:04:31,  4.37s/it]
  0%|          | 104/34750 [05:41<43:20:27,  4.50s/it]
  0%|          | 105/34750 [05:45<43:33:06,  4.53s/it]
  0%|          | 106/34750 [05:50<43:01:31,  4.47s/it]
  0%|          | 107/34750 [05:54<42:35:58,  4.43s/it]
  0%|          | 108/34750 [05:58<41:36:32,  4.32s/it]
  0%|          | 109/34750 [06:02<40:49:12,  4.24s/it]
  0%|          | 110/34750 [06:06<40:24:00,  4.20s/it]
  0%|          | 111/34750 [06:10<39:21:11,  4.09s/it]
  0%|          | 112/34750 [06:14<38:20:20,  3.98s/it]
  0%|          | 113/34750 [06:18<37:45:38,  3.92s/it]
  0%|          | 114/34750 [06:21<36:57:18,  3.84s/it]
  0%|          | 115/34750 [06:25<36:23:40,  3.78s/it]
  0%|          | 116/34750 [06:28<35:34:12,  3.70s/it]
  0%|          | 117/34750 [06:32<34:43:35,  3.61s/it]
  0%|          | 118/34750 [06:35<33:56:27,  3.53s/it]
  0%|          | 119/34750 [06:38<33:21:27,  3.47s/it]
  0%|          | 120/34750 [06:42<32:59:05,  3.43s/it]
  0%|          | 121/34750 [06:45<32:17:52,  3.36s/it]
  0%|          | 122/34750 [06:48<31:38:10,  3.29s/it]
  0%|          | 123/34750 [06:51<31:05:15,  3.23s/it]
  0%|          | 124/34750 [06:54<30:46:22,  3.20s/it]
  0%|          | 125/34750 [06:57<30:10:54,  3.14s/it]
  0%|          | 126/34750 [07:00<29:41:42,  3.09s/it]
  0%|          | 127/34750 [07:03<29:03:34,  3.02s/it]
  0%|          | 128/34750 [07:06<28:35:44,  2.97s/it]
  0%|          | 129/34750 [07:09<28:00:19,  2.91s/it]
  0%|          | 130/34750 [07:12<27:21:27,  2.84s/it]
  0%|          | 131/34750 [07:14<27:10:16,  2.83s/it]
  0%|          | 132/34750 [07:17<26:33:47,  2.76s/it]
  0%|          | 133/34750 [07:20<26:14:29,  2.73s/it]
  0%|          | 134/34750 [07:22<25:45:14,  2.68s/it]
  0%|          | 135/34750 [07:25<25:17:13,  2.63s/it]
  0%|          | 136/34750 [07:27<24:57:13,  2.60s/it]
  0%|          | 137/34750 [07:30<24:24:51,  2.54s/it]
  0%|          | 138/34750 [07:32<23:51:55,  2.48s/it]
  0%|          | 139/34750 [07:34<23:21:09,  2.43s/it]
  0%|          | 140/34750 [07:36<22:52:59,  2.38s/it]
  0%|          | 141/34750 [07:39<22:43:48,  2.36s/it]
  0%|          | 142/34750 [07:41<22:20:24,  2.32s/it]
  0%|          | 143/34750 [07:43<22:04:04,  2.30s/it]
  0%|          | 144/34750 [07:45<21:37:56,  2.25s/it]
  0%|          | 145/34750 [07:47<21:03:48,  2.19s/it]
  0%|          | 146/34750 [07:50<20:40:12,  2.15s/it]
  0%|          | 147/34750 [07:51<20:03:46,  2.09s/it]
  0%|          | 148/34750 [07:53<19:25:14,  2.02s/it]
  0%|          | 149/34750 [07:55<18:55:40,  1.97s/it]
  0%|          | 150/34750 [07:57<18:20:39,  1.91s/it]
  0%|          | 151/34750 [08:03<31:24:03,  3.27s/it]
  0%|          | 152/34750 [08:09<37:53:01,  3.94s/it]
  0%|          | 153/34750 [08:14<41:53:08,  4.36s/it]
  0%|          | 154/34750 [08:19<43:48:48,  4.56s/it]
  0%|          | 155/34750 [08:24<43:58:26,  4.58s/it]
  0%|          | 156/34750 [08:28<43:45:25,  4.55s/it]
  0%|          | 157/34750 [08:33<43:43:02,  4.55s/it]
  0%|          | 158/34750 [08:37<42:58:24,  4.47s/it]
  0%|          | 159/34750 [08:41<41:59:46,  4.37s/it]
  0%|          | 160/34750 [08:45<40:59:58,  4.27s/it]
  0%|          | 161/34750 [08:49<40:07:34,  4.18s/it]
  0%|          | 162/34750 [08:53<39:31:01,  4.11s/it]
  0%|          | 163/34750 [08:57<38:50:29,  4.04s/it]
  0%|          | 164/34750 [09:01<38:04:36,  3.96s/it]
  0%|          | 165/34750 [09:05<37:16:49,  3.88s/it]
  0%|          | 166/34750 [09:08<36:26:52,  3.79s/it]
  0%|          | 167/34750 [09:12<35:48:44,  3.73s/it]
  0%|          | 168/34750 [09:15<35:16:30,  3.67s/it]
  0%|          | 169/34750 [09:19<34:42:11,  3.61s/it]
  0%|          | 170/34750 [09:22<33:58:16,  3.54s/it]
  0%|          | 171/34750 [09:25<33:07:26,  3.45s/it]
  0%|          | 172/34750 [09:29<32:20:46,  3.37s/it]
  0%|          | 173/34750 [09:32<31:43:56,  3.30s/it]
  1%|          | 174/34750 [09:35<31:16:09,  3.26s/it]
  1%|          | 175/34750 [09:38<31:01:53,  3.23s/it]
  1%|          | 176/34750 [09:41<30:33:05,  3.18s/it]
  1%|          | 177/34750 [09:44<30:02:14,  3.13s/it]
  1%|          | 178/34750 [09:47<29:31:44,  3.07s/it]
  1%|          | 179/34750 [09:50<28:51:20,  3.00s/it]
  1%|          | 180/34750 [09:53<28:07:59,  2.93s/it]
  1%|          | 181/34750 [09:55<27:42:48,  2.89s/it]
  1%|          | 182/34750 [09:58<27:05:20,  2.82s/it]
  1%|          | 183/34750 [10:01<26:42:02,  2.78s/it]
  1%|          | 184/34750 [10:03<26:09:40,  2.72s/it]
  1%|          | 185/34750 [10:06<25:42:04,  2.68s/it]
  1%|          | 186/34750 [10:08<25:13:14,  2.63s/it]
  1%|          | 187/34750 [10:11<24:58:54,  2.60s/it]
  1%|          | 188/34750 [10:13<24:31:39,  2.55s/it]
  1%|          | 189/34750 [10:16<23:54:11,  2.49s/it]
  1%|          | 190/34750 [10:18<23:19:20,  2.43s/it]
  1%|          | 191/34750 [10:20<22:53:19,  2.38s/it]
  1%|          | 192/34750 [10:23<22:30:07,  2.34s/it]
  1%|          | 193/34750 [10:25<21:58:47,  2.29s/it]
  1%|          | 194/34750 [10:27<21:25:10,  2.23s/it]
  1%|          | 195/34750 [10:29<20:52:07,  2.17s/it]
  1%|          | 196/34750 [10:31<20:21:29,  2.12s/it]
  1%|          | 197/34750 [10:33<19:45:12,  2.06s/it]
  1%|          | 198/34750 [10:35<19:10:57,  2.00s/it]
  1%|          | 199/34750 [10:36<18:35:35,  1.94s/it]
  1%|          | 200/34750 [10:38<17:58:57,  1.87s/it]
                                                      
  1%|          | 200/34750 [10:38<17:58:57,  1.87s/it]
  1%|          | 201/34750 [10:44<30:29:43,  3.18s/it]
  1%|          | 202/34750 [10:50<36:43:22,  3.83s/it]
  1%|          | 203/34750 [10:55<40:17:17,  4.20s/it]
  1%|          | 204/34750 [11:00<41:57:33,  4.37s/it]
  1%|          | 205/34750 [11:04<42:29:54,  4.43s/it]
  1%|          | 206/34750 [11:09<42:44:42,  4.45s/it]
  1%|          | 207/34750 [11:13<41:56:02,  4.37s/it]
  1%|          | 208/34750 [11:17<41:20:03,  4.31s/it]
  1%|          | 209/34750 [11:21<40:36:04,  4.23s/it]
  1%|          | 210/34750 [11:25<40:15:07,  4.20s/it]
  1%|          | 211/34750 [11:29<39:13:51,  4.09s/it]
  1%|          | 212/34750 [11:33<38:18:15,  3.99s/it]
  1%|          | 213/34750 [11:37<37:31:08,  3.91s/it]
  1%|          | 214/34750 [11:40<36:50:20,  3.84s/it]
  1%|          | 215/34750 [11:44<36:03:03,  3.76s/it]
  1%|          | 216/34750 [11:47<35:19:22,  3.68s/it]
  1%|          | 217/34750 [11:51<34:49:41,  3.63s/it]
  1%|          | 218/34750 [11:54<33:58:09,  3.54s/it]
  1%|          | 219/34750 [11:57<33:17:16,  3.47s/it]
  1%|          | 220/34750 [12:01<32:42:24,  3.41s/it]
  1%|          | 221/34750 [12:04<32:12:04,  3.36s/it]
  1%|          | 222/34750 [12:07<31:21:23,  3.27s/it]
  1%|          | 223/34750 [12:10<30:51:33,  3.22s/it]
  1%|          | 224/34750 [12:13<30:13:54,  3.15s/it]
  1%|          | 225/34750 [12:16<29:45:11,  3.10s/it]
  1%|          | 226/34750 [12:19<29:23:35,  3.06s/it]
  1%|          | 227/34750 [12:22<28:42:32,  2.99s/it]
  1%|          | 228/34750 [12:25<28:22:04,  2.96s/it]
  1%|          | 229/34750 [12:28<27:48:51,  2.90s/it]
  1%|          | 230/34750 [12:30<27:11:50,  2.84s/it]
  1%|          | 231/34750 [12:33<26:35:09,  2.77s/it]
  1%|          | 232/34750 [12:35<26:10:51,  2.73s/it]
  1%|          | 233/34750 [12:38<25:52:41,  2.70s/it]
  1%|          | 234/34750 [12:41<25:20:57,  2.64s/it]
  1%|          | 235/34750 [12:43<24:56:17,  2.60s/it]
  1%|          | 236/34750 [12:45<24:12:05,  2.52s/it]
  1%|          | 237/34750 [12:48<23:36:45,  2.46s/it]
  1%|          | 238/34750 [12:50<23:15:05,  2.43s/it]
  1%|          | 239/34750 [12:52<22:55:05,  2.39s/it]
  1%|          | 240/34750 [12:55<22:24:01,  2.34s/it]
  1%|          | 241/34750 [12:57<22:06:03,  2.31s/it]
  1%|          | 242/34750 [12:59<21:38:36,  2.26s/it]
  1%|          | 243/34750 [13:01<21:15:14,  2.22s/it]
  1%|          | 244/34750 [13:03<20:47:51,  2.17s/it]
  1%|          | 245/34750 [13:05<20:23:43,  2.13s/it]
  1%|          | 246/34750 [13:07<20:17:09,  2.12s/it]
  1%|          | 247/34750 [13:09<19:45:30,  2.06s/it]
  1%|          | 248/34750 [13:11<19:07:56,  2.00s/it]
  1%|          | 249/34750 [13:13<18:36:53,  1.94s/it]
  1%|          | 250/34750 [13:15<17:51:35,  1.86s/it]
  1%|          | 251/34750 [13:21<31:21:56,  3.27s/it]
  1%|          | 252/34750 [13:27<37:56:29,  3.96s/it]
  1%|          | 253/34750 [13:32<41:27:01,  4.33s/it]
  1%|          | 254/34750 [13:37<43:20:41,  4.52s/it]
  1%|          | 255/34750 [13:42<43:46:29,  4.57s/it]
  1%|          | 256/34750 [13:46<44:07:24,  4.60s/it]
  1%|          | 257/34750 [13:51<43:48:40,  4.57s/it]
  1%|          | 258/34750 [13:55<42:57:47,  4.48s/it]
  1%|          | 259/34750 [13:59<42:16:34,  4.41s/it]
  1%|          | 260/34750 [14:03<41:36:00,  4.34s/it]
  1%|          | 261/34750 [14:07<40:22:55,  4.22s/it]
  1%|          | 262/34750 [14:11<39:34:28,  4.13s/it]
  1%|          | 263/34750 [14:15<38:38:45,  4.03s/it]
  1%|          | 264/34750 [14:19<37:37:21,  3.93s/it]
  1%|          | 265/34750 [14:23<37:05:51,  3.87s/it]
  1%|          | 266/34750 [14:26<36:12:42,  3.78s/it]
  1%|          | 267/34750 [14:30<35:15:32,  3.68s/it]
  1%|          | 268/34750 [14:33<34:42:08,  3.62s/it]
  1%|          | 269/34750 [14:36<34:05:26,  3.56s/it]
  1%|          | 270/34750 [14:40<33:17:54,  3.48s/it]
  1%|          | 271/34750 [14:43<32:24:37,  3.38s/it]
  1%|          | 272/34750 [14:46<31:57:35,  3.34s/it]
  1%|          | 273/34750 [14:49<31:33:17,  3.29s/it]
  1%|          | 274/34750 [14:52<31:13:36,  3.26s/it]
  1%|          | 275/34750 [14:56<30:33:40,  3.19s/it]
  1%|          | 276/34750 [14:58<29:57:52,  3.13s/it]
  1%|          | 277/34750 [15:01<29:24:24,  3.07s/it]
  1%|          | 278/34750 [15:04<28:49:51,  3.01s/it]
  1%|          | 279/34750 [15:07<28:08:27,  2.94s/it]
  1%|          | 280/34750 [15:10<27:22:33,  2.86s/it]
  1%|          | 281/34750 [15:12<26:57:28,  2.82s/it]
  1%|          | 282/34750 [15:15<26:27:44,  2.76s/it]
  1%|          | 283/34750 [15:18<26:13:36,  2.74s/it]
  1%|          | 284/34750 [15:21<26:13:15,  2.74s/it]
  1%|          | 285/34750 [15:23<25:57:40,  2.71s/it]
  1%|          | 286/34750 [15:26<25:28:50,  2.66s/it]
  1%|          | 287/34750 [15:28<25:04:44,  2.62s/it]
  1%|          | 288/34750 [15:31<24:26:40,  2.55s/it]
  1%|          | 289/34750 [15:33<23:41:08,  2.47s/it]
  1%|          | 290/34750 [15:35<23:22:38,  2.44s/it]
  1%|          | 291/34750 [15:38<22:57:12,  2.40s/it]
  1%|          | 292/34750 [15:40<22:17:49,  2.33s/it]
  1%|          | 293/34750 [15:42<21:41:20,  2.27s/it]
  1%|          | 294/34750 [15:44<21:08:58,  2.21s/it]
  1%|          | 295/34750 [15:46<20:50:26,  2.18s/it]
  1%|          | 296/34750 [15:48<20:32:20,  2.15s/it]
  1%|          | 297/34750 [15:50<19:59:15,  2.09s/it]
  1%|          | 298/34750 [15:52<19:26:06,  2.03s/it]
  1%|          | 299/34750 [15:54<18:49:19,  1.97s/it]
  1%|          | 300/34750 [15:56<18:10:12,  1.90s/it]
                                                      
  1%|          | 300/34750 [15:56<18:10:12,  1.90s/it]
  1%|          | 301/34750 [16:02<31:58:03,  3.34s/it]
  1%|          | 302/34750 [16:08<37:55:08,  3.96s/it]
  1%|          | 303/34750 [16:13<41:36:50,  4.35s/it]
  1%|          | 304/34750 [16:18<43:17:31,  4.52s/it]
  1%|          | 305/34750 [16:23<43:51:54,  4.58s/it]
  1%|          | 306/34750 [16:27<43:56:28,  4.59s/it]
  1%|          | 307/34750 [16:32<43:43:04,  4.57s/it]
  1%|          | 308/34750 [16:36<42:48:58,  4.48s/it]
  1%|          | 309/34750 [16:40<42:16:02,  4.42s/it]
  1%|          | 310/34750 [16:44<41:16:33,  4.31s/it]
  1%|          | 311/34750 [16:48<40:43:30,  4.26s/it]
  1%|          | 312/34750 [16:52<39:52:16,  4.17s/it]
  1%|          | 313/34750 [16:56<38:22:24,  4.01s/it]
  1%|          | 314/34750 [17:00<37:28:22,  3.92s/it]
  1%|          | 315/34750 [17:03<36:46:54,  3.85s/it]
  1%|          | 316/34750 [17:07<35:58:05,  3.76s/it]
  1%|          | 317/34750 [17:10<35:05:57,  3.67s/it]
  1%|          | 318/34750 [17:14<34:07:17,  3.57s/it]
  1%|          | 319/34750 [17:17<33:31:36,  3.51s/it]
  1%|          | 320/34750 [17:20<32:47:49,  3.43s/it]
  1%|          | 321/34750 [17:24<32:01:42,  3.35s/it]
  1%|          | 322/34750 [17:27<31:36:22,  3.30s/it]
  1%|          | 323/34750 [17:30<31:07:44,  3.26s/it]
  1%|          | 324/34750 [17:33<30:53:20,  3.23s/it]
  1%|          | 325/34750 [17:36<30:26:15,  3.18s/it]
  1%|          | 326/34750 [17:39<29:47:46,  3.12s/it]
  1%|          | 327/34750 [17:42<29:12:55,  3.06s/it]
  1%|          | 328/34750 [17:45<28:47:06,  3.01s/it]
  1%|          | 329/34750 [17:48<28:08:18,  2.94s/it]
  1%|          | 330/34750 [17:50<27:36:09,  2.89s/it]
  1%|          | 331/34750 [17:53<27:21:46,  2.86s/it]
  1%|          | 332/34750 [17:56<27:03:08,  2.83s/it]
  1%|          | 333/34750 [17:59<26:20:26,  2.76s/it]
  1%|          | 334/34750 [18:01<26:12:57,  2.74s/it]
  1%|          | 335/34750 [18:04<25:41:28,  2.69s/it]
  1%|          | 336/34750 [18:06<25:12:54,  2.64s/it]
  1%|          | 337/34750 [18:09<24:41:53,  2.58s/it]
  1%|          | 338/34750 [18:11<24:12:43,  2.53s/it]
  1%|          | 339/34750 [18:14<23:30:50,  2.46s/it]
  1%|          | 340/34750 [18:16<22:57:39,  2.40s/it]
  1%|          | 341/34750 [18:18<22:20:00,  2.34s/it]
  1%|          | 342/34750 [18:20<22:25:17,  2.35s/it]
  1%|          | 343/34750 [18:23<22:01:56,  2.31s/it]
  1%|          | 344/34750 [18:25<21:44:05,  2.27s/it]
  1%|          | 345/34750 [18:27<21:18:47,  2.23s/it]
  1%|          | 346/34750 [18:29<20:40:52,  2.16s/it]
  1%|          | 347/34750 [18:31<20:05:18,  2.10s/it]
  1%|          | 348/34750 [18:33<19:18:53,  2.02s/it]
  1%|          | 349/34750 [18:34<18:30:11,  1.94s/it]
  1%|          | 350/34750 [18:36<17:49:48,  1.87s/it]
  1%|          | 351/34750 [18:43<30:43:41,  3.22s/it]
  1%|          | 352/34750 [18:48<36:25:36,  3.81s/it]
  1%|          | 353/34750 [18:53<40:38:35,  4.25s/it]
  1%|          | 354/34750 [18:58<42:21:14,  4.43s/it]
  1%|          | 355/34750 [19:03<43:07:44,  4.51s/it]
  1%|          | 356/34750 [19:07<42:26:41,  4.44s/it]
  1%|          | 357/34750 [19:11<41:53:53,  4.39s/it]
  1%|          | 358/34750 [19:15<40:39:51,  4.26s/it]
  1%|          | 359/34750 [19:19<40:16:11,  4.22s/it]
  1%|          | 360/34750 [19:23<39:41:45,  4.16s/it]
  1%|          | 361/34750 [19:27<38:47:40,  4.06s/it]
  1%|          | 362/34750 [19:31<38:03:09,  3.98s/it]
  1%|          | 363/34750 [19:35<37:17:42,  3.90s/it]
  1%|          | 364/34750 [19:38<36:50:26,  3.86s/it]
  1%|          | 365/34750 [19:42<36:07:43,  3.78s/it]
  1%|          | 366/34750 [19:45<35:02:58,  3.67s/it]
  1%|          | 367/34750 [19:49<34:07:02,  3.57s/it]
  1%|          | 368/34750 [19:52<33:13:38,  3.48s/it]
  1%|          | 369/34750 [19:55<32:49:53,  3.44s/it]
  1%|          | 370/34750 [19:58<32:19:00,  3.38s/it]
  1%|          | 371/34750 [20:02<31:58:00,  3.35s/it]
  1%|          | 372/34750 [20:05<31:12:15,  3.27s/it]
  1%|          | 373/34750 [20:08<30:34:34,  3.20s/it]
  1%|          | 374/34750 [20:11<30:28:31,  3.19s/it]
  1%|          | 375/34750 [20:14<29:47:15,  3.12s/it]
  1%|          | 376/34750 [20:17<29:13:07,  3.06s/it]
  1%|          | 377/34750 [20:20<28:44:15,  3.01s/it]
  1%|          | 378/34750 [20:23<28:26:34,  2.98s/it]
  1%|          | 379/34750 [20:26<27:56:33,  2.93s/it]
  1%|          | 380/34750 [20:28<27:25:39,  2.87s/it]
  1%|          | 381/34750 [20:31<26:59:30,  2.83s/it]
  1%|          | 382/34750 [20:34<26:29:51,  2.78s/it]
  1%|          | 383/34750 [20:36<25:56:46,  2.72s/it]
  1%|          | 384/34750 [20:39<25:35:53,  2.68s/it]
  1%|          | 385/34750 [20:41<25:02:32,  2.62s/it]
  1%|          | 386/34750 [20:44<24:39:09,  2.58s/it]
  1%|          | 387/34750 [20:46<24:21:30,  2.55s/it]
  1%|          | 388/34750 [20:49<23:38:31,  2.48s/it]
  1%|          | 389/34750 [20:51<23:10:07,  2.43s/it]
  1%|          | 390/34750 [20:53<22:38:55,  2.37s/it]
  1%|          | 391/34750 [20:55<22:09:02,  2.32s/it]
  1%|          | 392/34750 [20:58<21:49:26,  2.29s/it]
  1%|          | 393/34750 [21:00<21:26:35,  2.25s/it]
  1%|          | 394/34750 [21:02<21:03:47,  2.21s/it]
  1%|          | 395/34750 [21:04<20:38:57,  2.16s/it]
  1%|          | 396/34750 [21:06<20:07:32,  2.11s/it]
  1%|          | 397/34750 [21:08<19:41:03,  2.06s/it]
  1%|          | 398/34750 [21:10<19:14:02,  2.02s/it]
  1%|          | 399/34750 [21:12<18:36:22,  1.95s/it]
  1%|          | 400/34750 [21:13<17:49:36,  1.87s/it]
                                                      
  1%|          | 400/34750 [21:13<17:49:36,  1.87s/it]
  1%|          | 401/34750 [21:20<31:30:35,  3.30s/it]
  1%|          | 402/34750 [21:26<38:21:07,  4.02s/it]
  1%|          | 403/34750 [21:31<41:14:36,  4.32s/it]
  1%|          | 404/34750 [21:35<42:39:12,  4.47s/it]
  1%|          | 405/34750 [21:40<43:39:49,  4.58s/it]
  1%|          | 406/34750 [21:45<44:18:50,  4.65s/it]
  1%|          | 407/34750 [21:49<43:34:28,  4.57s/it]
  1%|          | 408/34750 [21:54<42:43:07,  4.48s/it]
  1%|          | 409/34750 [21:58<42:01:38,  4.41s/it]
  1%|          | 410/34750 [22:02<41:12:36,  4.32s/it]
  1%|          | 411/34750 [22:06<40:08:54,  4.21s/it]
  1%|          | 412/34750 [22:10<39:03:22,  4.09s/it]
  1%|          | 413/34750 [22:14<38:34:32,  4.04s/it]
  1%|          | 414/34750 [22:17<37:40:02,  3.95s/it]
  1%|          | 415/34750 [22:21<36:37:15,  3.84s/it]
  1%|          | 416/34750 [22:25<36:12:41,  3.80s/it]
  1%|          | 417/34750 [22:28<35:24:02,  3.71s/it]
  1%|          | 418/34750 [22:32<34:40:10,  3.64s/it]
  1%|          | 419/34750 [22:35<33:57:25,  3.56s/it]
  1%|          | 420/34750 [22:38<33:08:40,  3.48s/it]
  1%|          | 421/34750 [22:42<32:18:45,  3.39s/it]
  1%|          | 422/34750 [22:45<31:30:05,  3.30s/it]
  1%|          | 423/34750 [22:48<30:52:25,  3.24s/it]
  1%|          | 424/34750 [22:51<30:20:07,  3.18s/it]
  1%|          | 425/34750 [22:54<30:16:08,  3.17s/it]
  1%|          | 426/34750 [22:57<29:37:23,  3.11s/it]
  1%|          | 427/34750 [23:00<29:02:17,  3.05s/it]
  1%|          | 428/34750 [23:03<28:32:26,  2.99s/it]
  1%|          | 429/34750 [23:05<27:55:36,  2.93s/it]
  1%|          | 430/34750 [23:08<27:22:38,  2.87s/it]
  1%|          | 431/34750 [23:11<26:42:52,  2.80s/it]
  1%|          | 432/34750 [23:13<26:09:53,  2.74s/it]
  1%|          | 433/34750 [23:16<25:33:50,  2.68s/it]
  1%|          | 434/34750 [23:19<25:26:14,  2.67s/it]
  1%|▏         | 435/34750 [23:21<25:08:01,  2.64s/it]
  1%|▏         | 436/34750 [23:24<24:28:11,  2.57s/it]
  1%|▏         | 437/34750 [23:26<23:45:04,  2.49s/it]
  1%|▏         | 438/34750 [23:28<23:09:23,  2.43s/it]
  1%|▏         | 439/34750 [23:30<22:35:50,  2.37s/it]
  1%|▏         | 440/34750 [23:33<22:09:21,  2.32s/it]
  1%|▏         | 441/34750 [23:35<21:32:40,  2.26s/it]
  1%|▏         | 442/34750 [23:37<21:02:35,  2.21s/it]
  1%|▏         | 443/34750 [23:39<20:37:00,  2.16s/it]
  1%|▏         | 444/34750 [23:41<20:17:59,  2.13s/it]
  1%|▏         | 445/34750 [23:43<20:03:47,  2.11s/it]
  1%|▏         | 446/34750 [23:45<19:49:33,  2.08s/it]
  1%|▏         | 447/34750 [23:47<19:19:48,  2.03s/it]
  1%|▏         | 448/34750 [23:49<18:48:10,  1.97s/it]
  1%|▏         | 449/34750 [23:51<18:28:39,  1.94s/it]
  1%|▏         | 450/34750 [23:52<17:45:13,  1.86s/it]
  1%|▏         | 451/34750 [23:59<30:08:37,  3.16s/it]
  1%|▏         | 452/34750 [24:04<37:08:51,  3.90s/it]
  1%|▏         | 453/34750 [24:09<40:16:04,  4.23s/it]
  1%|▏         | 454/34750 [24:14<41:18:56,  4.34s/it]
  1%|▏         | 455/34750 [24:18<42:06:45,  4.42s/it]
  1%|▏         | 456/34750 [24:23<41:51:37,  4.39s/it]
  1%|▏         | 457/34750 [24:27<41:49:04,  4.39s/it]
  1%|▏         | 458/34750 [24:31<41:03:14,  4.31s/it]
  1%|▏         | 459/34750 [24:35<40:29:08,  4.25s/it]
  1%|▏         | 460/34750 [24:39<39:39:10,  4.16s/it]
  1%|▏         | 461/34750 [24:43<38:56:37,  4.09s/it]
  1%|▏         | 462/34750 [24:47<38:30:33,  4.04s/it]
  1%|▏         | 463/34750 [24:51<37:55:04,  3.98s/it]
  1%|▏         | 464/34750 [24:55<37:01:56,  3.89s/it]
  1%|▏         | 465/34750 [24:58<36:01:26,  3.78s/it]
  1%|▏         | 466/34750 [25:02<35:06:25,  3.69s/it]
  1%|▏         | 467/34750 [25:05<34:18:13,  3.60s/it]
  1%|▏         | 468/34750 [25:08<33:47:41,  3.55s/it]
  1%|▏         | 469/34750 [25:12<32:59:07,  3.46s/it]
  1%|▏         | 470/34750 [25:15<32:19:51,  3.40s/it]
  1%|▏         | 471/34750 [25:18<31:52:25,  3.35s/it]
  1%|▏         | 472/34750 [25:21<31:32:56,  3.31s/it]
  1%|▏         | 473/34750 [25:24<30:51:50,  3.24s/it]
  1%|▏         | 474/34750 [25:28<30:17:54,  3.18s/it]
  1%|▏         | 475/34750 [25:31<29:52:40,  3.14s/it]
  1%|▏         | 476/34750 [25:34<29:21:56,  3.08s/it]
  1%|▏         | 477/34750 [25:37<29:08:35,  3.06s/it]
  1%|▏         | 478/34750 [25:39<28:25:03,  2.99s/it]
  1%|▏         | 479/34750 [25:42<28:15:33,  2.97s/it]
  1%|▏         | 480/34750 [25:45<27:32:19,  2.89s/it]
  1%|▏         | 481/34750 [25:48<27:02:24,  2.84s/it]
  1%|▏         | 482/34750 [25:50<26:38:29,  2.80s/it]
  1%|▏         | 483/34750 [25:53<26:09:21,  2.75s/it]
  1%|▏         | 484/34750 [25:56<26:07:06,  2.74s/it]
  1%|▏         | 485/34750 [25:58<25:46:30,  2.71s/it]
  1%|▏         | 486/34750 [26:01<25:11:05,  2.65s/it]
  1%|▏         | 487/34750 [26:03<24:41:17,  2.59s/it]
  1%|▏         | 488/34750 [26:06<24:11:10,  2.54s/it]
  1%|▏         | 489/34750 [26:08<23:45:21,  2.50s/it]
  1%|▏         | 490/34750 [26:10<23:12:34,  2.44s/it]
  1%|▏         | 491/34750 [26:13<22:40:29,  2.38s/it]
  1%|▏         | 492/34750 [26:15<22:19:29,  2.35s/it]
  1%|▏         | 493/34750 [26:17<22:03:05,  2.32s/it]
  1%|▏         | 494/34750 [26:19<21:32:01,  2.26s/it]
  1%|▏         | 495/34750 [26:21<21:00:30,  2.21s/it]
  1%|▏         | 496/34750 [26:23<20:25:39,  2.15s/it]
  1%|▏         | 497/34750 [26:25<19:55:06,  2.09s/it]
  1%|▏         | 498/34750 [26:27<19:13:21,  2.02s/it]
  1%|▏         | 499/34750 [26:29<18:38:20,  1.96s/it]
  1%|▏         | 500/34750 [26:31<17:50:56,  1.88s/it]
                                                      
  1%|▏         | 500/34750 [26:31<17:50:56,  1.88s/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 = 456
+  Batch size = 8
+{'loss': 47.2908, 'learning_rate': 3.675e-06, 'epoch': 0.14}
+{'loss': 33.9125, 'learning_rate': 7.425e-06, 'epoch': 0.29}
+{'loss': 26.6068, 'learning_rate': 1.1174999999999999e-05, 'epoch': 0.43}
+{'loss': 23.2775, 'learning_rate': 1.4925e-05, 'epoch': 0.57}
+{'loss': 19.7138, 'learning_rate': 1.8675e-05, 'epoch': 0.72}
+
+
  0%|          | 0/57 [00:00<?, ?it/s]
+
  4%|▎         | 2/57 [00:00<00:19,  2.82it/s]
+
  5%|▌         | 3/57 [00:01<00:25,  2.12it/s]
+
  7%|▋         | 4/57 [00:02<00:28,  1.83it/s]
+
  9%|▉         | 5/57 [00:02<00:28,  1.85it/s]
+
 11%|█         | 6/57 [00:03<00:28,  1.78it/s]
+
 12%|█▏        | 7/57 [00:03<00:28,  1.76it/s]
+
 14%|█▍        | 8/57 [00:04<00:28,  1.72it/s]
+
 16%|█▌        | 9/57 [00:04<00:27,  1.72it/s]
+
 18%|█▊        | 10/57 [00:05<00:26,  1.75it/s]
+
 19%|█▉        | 11/57 [00:06<00:27,  1.68it/s]
+
 21%|██        | 12/57 [00:06<00:29,  1.53it/s]
+
 23%|██▎       | 13/57 [00:07<00:31,  1.42it/s]
+
 25%|██▍       | 14/57 [00:08<00:29,  1.44it/s]
+
 26%|██▋       | 15/57 [00:09<00:31,  1.34it/s]
+
 28%|██▊       | 16/57 [00:09<00:28,  1.43it/s]
+
 30%|██▉       | 17/57 [00:10<00:26,  1.50it/s]
+
 32%|███▏      | 18/57 [00:11<00:24,  1.58it/s]
+
 33%|███▎      | 19/57 [00:11<00:23,  1.65it/s]
+
 35%|███▌      | 20/57 [00:12<00:22,  1.64it/s]
+
 37%|███▋      | 21/57 [00:12<00:21,  1.65it/s]
+
 39%|███▊      | 22/57 [00:13<00:23,  1.52it/s]
+
 40%|████      | 23/57 [00:14<00:24,  1.40it/s]
+
 42%|████▏     | 24/57 [00:15<00:23,  1.42it/s]
+
 44%|████▍     | 25/57 [00:15<00:21,  1.47it/s]
+
 46%|████▌     | 26/57 [00:16<00:19,  1.55it/s]
+
 47%|████▋     | 27/57 [00:16<00:18,  1.65it/s]
+
 49%|████▉     | 28/57 [00:17<00:18,  1.59it/s]
+
 51%|█████     | 29/57 [00:18<00:17,  1.57it/s]
+
 53%|█████▎    | 30/57 [00:18<00:15,  1.70it/s]
+
 54%|█████▍    | 31/57 [00:19<00:14,  1.83it/s]
+
 56%|█████▌    | 32/57 [00:19<00:14,  1.78it/s]
+
 58%|█████▊    | 33/57 [00:20<00:14,  1.66it/s]
+
 60%|█████▉    | 34/57 [00:20<00:14,  1.63it/s]
+
 61%|██████▏   | 35/57 [00:21<00:14,  1.56it/s]
+
 63%|██████▎   | 36/57 [00:22<00:13,  1.57it/s]
+
 65%|██████▍   | 37/57 [00:22<00:13,  1.53it/s]
+
 67%|██████▋   | 38/57 [00:23<00:12,  1.47it/s]
+
 68%|██████▊   | 39/57 [00:24<00:12,  1.46it/s]
+
 70%|███████   | 40/57 [00:25<00:11,  1.43it/s]
+
 72%|███████▏  | 41/57 [00:25<00:11,  1.37it/s]
+
 74%|███████▎  | 42/57 [00:26<00:11,  1.35it/s]
+
 75%|███████▌  | 43/57 [00:27<00:10,  1.39it/s]
+
 77%|███████▋  | 44/57 [00:28<00:09,  1.36it/s]
+
 79%|███████▉  | 45/57 [00:28<00:07,  1.54it/s]
+
 81%|████████  | 46/57 [00:29<00:06,  1.58it/s]
+
 82%|████████▏ | 47/57 [00:29<00:06,  1.55it/s]
+
 84%|████████▍ | 48/57 [00:30<00:05,  1.63it/s]
+
 86%|████████▌ | 49/57 [00:31<00:04,  1.67it/s]
+
 88%|████████▊ | 50/57 [00:31<00:04,  1.67it/s]
+
 89%|████████▉ | 51/57 [00:32<00:03,  1.66it/s]
+
 91%|█████████ | 52/57 [00:32<00:02,  1.69it/s]
+
 93%|█████████▎| 53/57 [00:33<00:02,  1.80it/s]
+
 95%|█████████▍| 54/57 [00:33<00:01,  1.77it/s]
+
 96%|█████████▋| 55/57 [00:34<00:01,  1.61it/s]
+
 98%|█████████▊| 56/57 [00:35<00:00,  1.56it/s]
+
                                                      
+
                                               
  1%|▏         | 500/34750 [27:12<17:50:56,  1.88s/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 sha256:95d95792c6081f64b83bc4a1b3c5f062d412822f9a3fa8e9f2bc12785dd1634d
+size 1266872433

run.sh

@@ -0,0 +1,34 @@
+python run_speech_recognition_ctc.py \
+	--dataset_name="kresnik/zeroth_korean" \
+	--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
+	--dataset_config_name="clean" \
+	--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="text" \
+	--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
+	--save_steps="500" \
+	--eval_steps="500" \
+	--logging_steps="100" \
+	--layerdrop="0.0" \
+	--activation_dropout="0.1" \
+	--save_total_limit="3" \
+	--freeze_feature_encoder \
+	--feat_proj_dropout="0.0" \
+	--mask_time_prob="0.75" \
+	--mask_time_length="10" \
+	--mask_feature_prob="0.25" \
+	--mask_feature_length="64" \
+	--gradient_checkpointing \
+	--use_auth_token \
+	--fp16 \
+	--group_by_length \
+	--do_train --do_eval \
+	--push_to_hub

run_speech_recognition_ctc.py

@@ -0,0 +1,829 @@
+#!/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",
+        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", "cer"],
+        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()

runs/Jan31_07-15-59_job-2c68f48a-2d5d-4013-9043-3f2cb25f3ff6/1643613501.488685/events.out.tfevents.1643613501.job-2c68f48a-2d5d-4013-9043-3f2cb25f3ff6.1151936.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28a6ab5fcdee80fd31c69dc157696d3063a1cc27099a2452f6695c51cba48628
+size 4753

runs/Jan31_07-15-59_job-2c68f48a-2d5d-4013-9043-3f2cb25f3ff6/events.out.tfevents.1643613501.job-2c68f48a-2d5d-4013-9043-3f2cb25f3ff6.1151936.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2899aba1a5c0264df3c5c6326a68747e153f39041d06ca0ad3d1b42b63f3b01b
+size 5833

special_tokens_map.json

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

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "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 sha256:4af57085f0712099c06c191a8e3123d5fbba4119a615b1c3ec5ef78e066139b2
+size 2991

vocab.json

@@ -0,0 +1 @@
+{"가": 1, "각": 2, "간": 3, "갈": 4, "감": 5, "갑": 6, "값": 7, "갓": 8, "갔": 9, "강": 10, "갖": 11, "같": 12, "갚": 13, "개": 14, "객": 15, "갠": 16, "갤": 17, "갯": 18, "갱": 19, "거": 20, "걱": 21, "건": 22, "걷": 23, "걸": 24, "검": 25, "겁": 26, "것": 27, "겉": 28, "게": 29, "겐": 30, "겟": 31, "겠": 32, "겨": 33, "격": 34, "겪": 35, "견": 36, "결": 37, "겸": 38, "겹": 39, "겼": 40, "경": 41, "곁": 42, "계": 43, "고": 44, "곡": 45, "곤": 46, "곧": 47, "골": 48, "곰": 49, "곱": 50, "곳": 51, "공": 52, "곶": 53, "과": 54, "곽": 55, "관": 56, "괄": 57, "괌": 58, "광": 59, "괘": 60, "괭": 61, "괴": 62, "굉": 63, "교": 64, "구": 65, "국": 66, "군": 67, "굳": 68, "굴": 69, "굵": 70, "굶": 71, "굽": 72, "굿": 73, "궁": 74, "궈": 75, "권": 76, "궐": 77, "궤": 78, "귀": 79, "규": 80, "균": 81, "그": 82, "극": 83, "근": 84, "글": 85, "금": 86, "급": 87, "긋": 88, "긍": 89, "기": 90, "긴": 91, "길": 92, "김": 93, "깁": 94, "깃": 95, "깊": 96, "까": 97, "깎": 98, "깔": 99, "깝": 100, "깡": 101, "깥": 102, "깨": 103, "깬": 104, "꺼": 105, "껍": 106, "껏": 107, "껑": 108, "께": 109, "껴": 110, "꼈": 111, "꼬": 112, "꼭": 113, "꼴": 114, "꼼": 115, "꼽": 116, "꽁": 117, "꽂": 118, "꽃": 119, "꽝": 120, "꽥": 121, "꾸": 122, "꾼": 123, "꿀": 124, "꿇": 125, "꿈": 126, "꿎": 127, "꿔": 128, "꿨": 129, "꿰": 130, "뀌": 131, "뀐": 132, "끄": 133, "끅": 134, "끈": 135, "끊": 136, "끌": 137, "끓": 138, "끔": 139, "끗": 140, "끝": 141, "끼": 142, "끽": 143, "낀": 144, "낌": 145, "나": 146, "낙": 147, "난": 148, "날": 149, "낡": 150, "남": 151, "납": 152, "낫": 153, "났": 154, "낭": 155, "낮": 156, "낯": 157, "낳": 158, "내": 159, "낸": 160, "낼": 161, "냄": 162, "냅": 163, "냇": 164, "냈": 165, "냉": 166, "냐": 167, "냥": 168, "너": 169, "넉": 170, "넌": 171, "널": 172, "넓": 173, "넘": 174, "넛": 175, "넣": 176, "네": 177, "넥": 178, "넨": 179, "넷": 180, "녀": 181, "녁": 182, "년": 183, "념": 184, "녔": 185, "녕": 186, "녘": 187, "노": 188, "녹": 189, "논": 190, "놀": 191, "놈": 192, "농": 193, "높": 194, "놓": 195, "놨": 196, "뇌": 197, "뇨": 198, "뇽": 199, "누": 200, "눈": 201, "눌": 202, "눠": 203, "뉜": 204, "뉴": 205, "늉": 206, "느": 207, "는": 208, "늘": 209, "늠": 210, "능": 211, "늦": 212, "늪": 213, "늬": 214, "니": 215, "닉": 216, "닌": 217, "닐": 218, "님": 219, "닙": 220, "닛": 221, "닝": 222, "다": 223, "닥": 224, "닦": 225, "단": 226, "닫": 227, "달": 228, "닭": 229, "닮": 230, "닳": 231, "담": 232, "답": 233, "닷": 234, "당": 235, "닿": 236, "대": 237, "댁": 238, "댐": 239, "댓": 240, "더": 241, "덕": 242, "던": 243, "덜": 244, "덟": 245, "덤": 246, "덧": 247, "덩": 248, "덫": 249, "덮": 250, "데": 251, "덴": 252, "델": 253, "뎅": 254, "뎌": 255, "도": 256, "독": 257, "돈": 258, "돋": 259, "돌": 260, "돔": 261, "돕": 262, "돗": 263, "동": 264, "돛": 265, "돼": 266, "됐": 267, "되": 268, "된": 269, "될": 270, "됨": 271, "됩": 272, "두": 273, "둑": 274, "둔": 275, "둘": 276, "둠": 277, "둥": 278, "둬": 279, "뒀": 280, "뒤": 281, "뒷": 282, "듀": 283, "드": 284, "득": 285, "든": 286, "듣": 287, "들": 288, "듬": 289, "듭": 290, "듯": 291, "등": 292, "디": 293, "딘": 294, "딜": 295, "딥": 296, "딨": 297, "딩": 298, "딪": 299, "따": 300, "딱": 301, "딴": 302, "딸": 303, "땀": 304, "땄": 305, "땅": 306, "때": 307, "땐": 308, "땠": 309, "땡": 310, "떠": 311, "떡": 312, "떤": 313, "떨": 314, "떴": 315, "떻": 316, "떼": 317, "뗄": 318, "또": 319, "똑": 320, "똥": 321, "뚜": 322, "뚝": 323, "뚫": 324, "뛰": 325, "뛴": 326, "뜨": 327, "뜯": 328, "뜸": 329, "뜻": 330, "띄": 331, "띔": 332, "띠": 333, "띤": 334, "띨": 335, "띵": 336, "라": 337, "락": 338, "란": 339, "랄": 340, "람": 341, "랍": 342, "랐": 343, "랑": 344, "래": 345, "랙": 346, "랜": 347, "램": 348, "랫": 349, "랬": 350, "랭": 351, "랴": 352, "략": 353, "량": 354, "러": 355, "럭": 356, "런": 357, "럴": 358, "럼": 359, "럽": 360, "렀": 361, "렁": 362, "렇": 363, "레": 364, "렉": 365, "렌": 366, "렘": 367, "렛": 368, "려": 369, "력": 370, "련": 371, "렬": 372, "렴": 373, "렵": 374, "렷": 375, "렸": 376, "령": 377, "례": 378, "로": 379, "록": 380, "론": 381, "롤": 382, "롬": 383, "롭": 384, "롯": 385, "롱": 386, "뢰": 387, "료": 388, "룡": 389, "루": 390, "룬": 391, "룰": 392, "룸": 393, "룹": 394, "뤄": 395, "뤘": 396, "뤼": 397, "류": 398, "륙": 399, "륜": 400, "률": 401, "륭": 402, "르": 403, "륵": 404, "른": 405, "를": 406, "름": 407, "릅": 408, "릇": 409, "릉": 410, "릎": 411, "리": 412, "릭": 413, "린": 414, "릴": 415, "림": 416, "립": 417, "릿": 418, "링": 419, "마": 420, "막": 421, "만": 422, "많": 423, "맏": 424, "말": 425, "맑": 426, "맘": 427, "맙": 428, "맛": 429, "망": 430, "맞": 431, "맡": 432, "매": 433, "맥": 434, "맨": 435, "맹": 436, "맺": 437, "머": 438, "먹": 439, "먼": 440, "멀": 441, "멈": 442, "멋": 443, "멍": 444, "메": 445, "멕": 446, "멘": 447, "멜": 448, "멧": 449, "며": 450, "면": 451, "멸": 452, "명": 453, "몇": 454, "모": 455, "목": 456, "몫": 457, "몬": 458, "몰": 459, "몸": 460, "못": 461, "몽": 462, "묘": 463, "무": 464, "묵": 465, "묶": 466, "문": 467, "묻": 468, "물": 469, "뭄": 470, "뭇": 471, "뭐": 472, "뭔": 473, "뭘": 474, "뮤": 475, "뮬": 476, "므": 477, "믈": 478, "미": 479, "믹": 480, "민": 481, "믿": 482, "밀": 483, "밋": 484, "밌": 485, "밍": 486, "및": 487, "밑": 488, "바": 489, "박": 490, "밖": 491, "반": 492, "받": 493, "발": 494, "밝": 495, "밤": 496, "밥": 497, "방": 498, "밭": 499, "배": 500, "백": 501, "밴": 502, "뱀": 503, "뱃": 504, "뱅": 505, "버": 506, "벅": 507, "번": 508, "벌": 509, "범": 510, "법": 511, "벗": 512, "벚": 513, "베": 514, "벤": 515, "벨": 516, "벳": 517, "벼": 518, "벽": 519, "변": 520, "별": 521, "볍": 522, "볐": 523, "병": 524, "볕": 525, "보": 526, "복": 527, "볶": 528, "본": 529, "볼": 530, "봄": 531, "봅": 532, "봇": 533, "봉": 534, "봐": 535, "봤": 536, "뵈": 537, "뵙": 538, "부": 539, "북": 540, "분": 541, "불": 542, "붉": 543, "붐": 544, "붓": 545, "붕": 546, "붙": 547, "뷔": 548, "뷰": 549, "브": 550, "븐": 551, "블": 552, "비": 553, "빅": 554, "빈": 555, "빌": 556, "빗": 557, "빙": 558, "빚": 559, "빛": 560, "빠": 561, "빨": 562, "빵": 563, "빼": 564, "뺀": 565, "뺌": 566, "뺏": 567, "뺑": 568, "뻐": 569, "뻑": 570, "뻔": 571, "뻗": 572, "뻘": 573, "뼈": 574, "뽀": 575, "뽑": 576, "뽕": 577, "뿌": 578, "뿐": 579, "뿜": 580, "쁘": 581, "쁜": 582, "쁩": 583, "삐": 584, "사": 585, "삭": 586, "산": 587, "살": 588, "삶": 589, "삼": 590, "삽": 591, "삿": 592, "샀": 593, "상": 594, "새": 595, "색": 596, "샌": 597, "샐": 598, "샘": 599, "생": 600, "샤": 601, "샬": 602, "샵": 603, "샷": 604, "서": 605, "석": 606, "섞": 607, "선": 608, "섣": 609, "설": 610, "섬": 611, "섭": 612, "섯": 613, "섰": 614, "성": 615, "세": 616, "섹": 617, "센": 618, "셀": 619, "셈": 620, "셉": 621, "셋": 622, "셔": 623, "션": 624, "셜": 625, "셨": 626, "셰": 627, "소": 628, "속": 629, "손": 630, "솔": 631, "솜": 632, "솟": 633, "송": 634, "솥": 635, "쇄": 636, "쇠": 637, "쇤": 638, "쇼": 639, "숍": 640, "수": 641, "숙": 642, "순": 643, "술": 644, "숨": 645, "숭": 646, "숲": 647, "쉬": 648, "쉰": 649, "쉼": 650, "쉽": 651, "슈": 652, "슐": 653, "스": 654, "슨": 655, "슬": 656, "슴": 657, "습": 658, "슷": 659, "승": 660, "시": 661, "식": 662, "신": 663, "실": 664, "싫": 665, "심": 666, "십": 667, "싱": 668, "싶": 669, "싸": 670, "싹": 671, "싼": 672, "쌀": 673, "쌈": 674, "쌌": 675, "쌍": 676, "쌓": 677, "써": 678, "썩": 679, "썰": 680, "썼": 681, "쏘": 682, "쏜": 683, "쏟": 684, "쏠": 685, "쑤": 686, "쓰": 687, "쓴": 688, "쓸": 689, "씀": 690, "씁": 691, "씌": 692, "씨": 693, "씩": 694, "씬": 695, "씸": 696, "씻": 697, "아": 698, "악": 699, "안": 700, "앉": 701, "않": 702, "알": 703, "앓": 704, "암": 705, "압": 706, "앗": 707, "았": 708, "앙": 709, "앞": 710, "애": 711, "액": 712, "앤": 713, "앨": 714, "앱": 715, "앵": 716, "야": 717, "약": 718, "얇": 719, "양": 720, "얗": 721, "얘": 722, "어": 723, "억": 724, "언": 725, "얹": 726, "얻": 727, "얼": 728, "얽": 729, "엄": 730, "업": 731, "없": 732, "엇": 733, "었": 734, "엉": 735, "엎": 736, "에": 737, "엑": 738, "엔": 739, "엘": 740, "엠": 741, "엡": 742, "엣": 743, "여": 744, "역": 745, "연": 746, "열": 747, "엷": 748, "염": 749, "엽": 750, "엿": 751, "였": 752, "영": 753, "옆": 754, "예": 755, "옛": 756, "오": 757, "옥": 758, "온": 759, "올": 760, "옮": 761, "옳": 762, "옴": 763, "옵": 764, "옷": 765, "옹": 766, "와": 767, "완": 768, "왈": 769, "왔": 770, "왕": 771, "왜": 772, "외": 773, "왼": 774, "요": 775, "욕": 776, "용": 777, "우": 778, "욱": 779, "운": 780, "울": 781, "움": 782, "웁": 783, "웃": 784, "웅": 785, "워": 786, "원": 787, "월": 788, "웠": 789, "웨": 790, "웬": 791, "웹": 792, "위": 793, "윈": 794, "윌": 795, "윗": 796, "윙": 797, "유": 798, "육": 799, "윤": 800, "율": 801, "융": 802, "으": 803, "은": 804, "을": 805, "음": 806, "읍": 807, "응": 808, "의": 809, "이": 810, "익": 811, "인": 812, "일": 813, "읽": 814, "잃": 815, "임": 816, "입": 817, "잇": 818, "있": 819, "잉": 820, "잊": 821, "잎": 822, "자": 823, "작": 824, "잔": 825, "잖": 826, "잘": 827, "잠": 828, "잡": 829, "잣": 830, "장": 831, "잦": 832, "재": 833, "잭": 834, "잰": 835, "잽": 836, "쟁": 837, "저": 838, "적": 839, "전": 840, "절": 841, "젊": 842, "젋": 843, "점": 844, "접": 845, "젓": 846, "정": 847, "젖": 848, "제": 849, "젝": 850, "젠": 851, "젤": 852, "져": 853, "젼": 854, "졌": 855, "조": 856, "족": 857, "존": 858, "졸": 859, "좀": 860, "좁": 861, "종": 862, "좋": 863, "좌": 864, "죄": 865, "죠": 866, "주": 867, "죽": 868, "준": 869, "줄": 870, "줍": 871, "중": 872, "줘": 873, "줬": 874, "쥐": 875, "쥔": 876, "쥘": 877, "쥬": 878, "즈": 879, "즉": 880, "즌": 881, "즐": 882, "즘": 883, "증": 884, "지": 885, "직": 886, "진": 887, "질": 888, "짊": 889, "짐": 890, "집": 891, "짓": 892, "징": 893, "짖": 894, "짙": 895, "짚": 896, "짜": 897, "짝": 898, "짧": 899, "짬": 900, "째": 901, "쨌": 902, "쩌": 903, "쩍": 904, "쩔": 905, "쩜": 906, "쪼": 907, "쪽": 908, "쫄": 909, "쫓": 910, "쭉": 911, "쯤": 912, "찌": 913, "찍": 914, "찔": 915, "찢": 916, "찧": 917, "차": 918, "착": 919, "찬": 920, "찮": 921, "찰": 922, "참": 923, "찻": 924, "창": 925, "찾": 926, "채": 927, "책": 928, "챌": 929, "챔": 930, "챙": 931, "챠": 932, "처": 933, "척": 934, "천": 935, "철": 936, "첨": 937, "첩": 938, "첫": 939, "청": 940, "체": 941, "첸": 942, "첼": 943, "쳇": 944, "쳐": 945, "쳤": 946, "초": 947, "촉": 948, "촌": 949, "촘": 950, "촛": 951, "총": 952, "촨": 953, "촬": 954, "최": 955, "추": 956, "축": 957, "춘": 958, "출": 959, "춤": 960, "충": 961, "춰": 962, "취": 963, "츠": 964, "측": 965, "츰": 966, "층": 967, "치": 968, "칙": 969, "친": 970, "칠": 971, "침": 972, "칩": 973, "칫": 974, "칭": 975, "카": 976, "칸": 977, "칼": 978, "캉": 979, "캐": 980, "캔": 981, "캘": 982, "캠": 983, "커": 984, "컥": 985, "컨": 986, "컫": 987, "컴": 988, "컵": 989, "컷": 990, "컸": 991, "케": 992, "켈": 993, "켐": 994, "켑": 995, "켓": 996, "켜": 997, "켰": 998, "코": 999, "콘": 1000, "콜": 1001, "콤": 1002, "콥": 1003, "콧": 1004, "콩": 1005, "쾌": 1006, "쿄": 1007, "쿠": 1008, "쿡": 1009, "쿨": 1010, "쿼": 1011, "퀴": 1012, "큐": 1013, "크": 1014, "큰": 1015, "클": 1016, "큼": 1017, "키": 1018, "킥": 1019, "킨": 1020, "킬": 1021, "킷": 1022, "킹": 1023, "타": 1024, "탁": 1025, "탄": 1026, "탈": 1027, "탐": 1028, "탑": 1029, "탓": 1030, "탕": 1031, "태": 1032, "택": 1033, "탠": 1034, "탬": 1035, "탱": 1036, "터": 1037, "턱": 1038, "턴": 1039, "털": 1040, "텃": 1041, "텅": 1042, "테": 1043, "텍": 1044, "텐": 1045, "텔": 1046, "템": 1047, "텼": 1048, "토": 1049, "톡": 1050, "톤": 1051, "톨": 1052, "톰": 1053, "통": 1054, "퇴": 1055, "투": 1056, "툴": 1057, "툼": 1058, "퉁": 1059, "튀": 1060, "튜": 1061, "튬": 1062, "트": 1063, "특": 1064, "튼": 1065, "튿": 1066, "틀": 1067, "틈": 1068, "티": 1069, "틱": 1070, "틴": 1071, "틸": 1072, "팀": 1073, "팅": 1074, "파": 1075, "팍": 1076, "팎": 1077, "판": 1078, "팔": 1079, "팜": 1080, "팡": 1081, "패": 1082, "팩": 1083, "팬": 1084, "팰": 1085, "팻": 1086, "팽": 1087, "퍼": 1088, "펀": 1089, "펄": 1090, "펌": 1091, "페": 1092, "펜": 1093, "펠": 1094, "펫": 1095, "펴": 1096, "편": 1097, "펼": 1098, "폄": 1099, "폈": 1100, "평": 1101, "폐": 1102, "포": 1103, "폭": 1104, "폰": 1105, "폴": 1106, "폼": 1107, "표": 1108, "푸": 1109, "푹": 1110, "푼": 1111, "풀": 1112, "품": 1113, "풋": 1114, "풍": 1115, "퓨": 1116, "퓰": 1117, "프": 1118, "픈": 1119, "플": 1120, "픔": 1121, "피": 1122, "픽": 1123, "핀": 1124, "필": 1125, "핏": 1126, "핑": 1127, "하": 1128, "학": 1129, "한": 1130, "할": 1131, "함": 1132, "합": 1133, "핫": 1134, "항": 1135, "해": 1136, "핵": 1137, "핸": 1138, "햇": 1139, "했": 1140, "행": 1141, "향": 1142, "허": 1143, "헌": 1144, "헐": 1145, "험": 1146, "헝": 1147, "헤": 1148, "헨": 1149, "헬": 1150, "헴": 1151, "헷": 1152, "혀": 1153, "혁": 1154, "현": 1155, "혈": 1156, "혐": 1157, "협": 1158, "혔": 1159, "형": 1160, "혜": 1161, "호": 1162, "혹": 1163, "혼": 1164, "홀": 1165, "홈": 1166, "홉": 1167, "홍": 1168, "화": 1169, "확": 1170, "환": 1171, "활": 1172, "황": 1173, "회": 1174, "획": 1175, "횡": 1176, "효": 1177, "후": 1178, "훈": 1179, "훌": 1180, "훤": 1181, "훨": 1182, "훼": 1183, "휘": 1184, "휩": 1185, "휴": 1186, "흉": 1187, "흐": 1188, "흑": 1189, "흔": 1190, "흘": 1191, "흙": 1192, "흠": 1193, "흡": 1194, "흥": 1195, "흩": 1196, "희": 1197, "흰": 1198, "히": 1199, "힌": 1200, "힐": 1201, "힘": 1202, "|": 0, "[UNK]": 1203, "[PAD]": 1204}