|
#!/usr/bin/env bash |
|
|
|
CUDA_VISIBLE_DEVICES="0" python run_speech_recognition_ctc.py \ |
|
--dataset_name="timit_asr" \ |
|
--model_name_or_path="ntu-spml/distilhubert" \ |
|
--overwrite_output_dir \ |
|
--output_dir="./distilhubert-timit" \ |
|
--train_split_name="train" \ |
|
--num_train_epochs="20" \ |
|
--per_device_train_batch_size="32" \ |
|
--per_device_eval_batch_size="1" \ |
|
--weight_decay="0.005" \ |
|
--learning_rate="1e-4" \ |
|
--warmup_steps="1000" \ |
|
--evaluation_strategy="steps" \ |
|
--text_column_name="text" \ |
|
--save_steps="400" \ |
|
--eval_steps="100" \ |
|
--logging_steps="10" \ |
|
--layerdrop="0.0" \ |
|
--save_total_limit="3" \ |
|
--freeze_feature_extractor \ |
|
--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \ |
|
--fp16 \ |
|
--group_by_length \ |
|
--push_to_hub \ |
|
--do_train --do_eval |
|
|