run.sh

#!/usr/bin/env bash
#	--gradient_checkpointing \
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