|
#!/usr/bin/env bash |
|
python -m torch.distributed.launch \ |
|
--nproc_per_node 8 run_speech_recognition_ctc.py \ |
|
--dataset_name="librispeech_asr" \ |
|
--model_name_or_path="microsoft/wavlm-base" \ |
|
--dataset_config_name="clean" \ |
|
--train_split_name="train.100" \ |
|
--eval_split_name="validation" \ |
|
--output_dir="./wavlm-libri-clean-100h-base" \ |
|
--preprocessing_num_workers="16" \ |
|
--length_column_name="input_length" \ |
|
--overwrite_output_dir \ |
|
--num_train_epochs="3" \ |
|
--per_device_train_batch_size="4" \ |
|
--per_device_eval_batch_size="4" \ |
|
--gradient_accumulation_steps="1" \ |
|
--learning_rate="3e-4" \ |
|
--warmup_steps="500" \ |
|
--evaluation_strategy="steps" \ |
|
--text_column_name="text" \ |
|
--save_steps="500" \ |
|
--eval_steps="300" \ |
|
--logging_steps="1" \ |
|
--layerdrop="0.0" \ |
|
--save_total_limit="3" \ |
|
--freeze_feature_extractor \ |
|
--gradient_checkpointing \ |
|
--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” \ |
|
--fp16 \ |
|
--group_by_length \ |
|
--do_train --do_eval |
