#!/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="asapp/sew-mid-100k" \
	--dataset_config_name="clean" \
	--train_split_name="train.100" \
	--eval_split_name="validation" \
	--output_dir="./sew-mid-100k-librispeech-clean-100h-ft" \
	--preprocessing_num_workers="16" \
	--overwrite_output_dir \
	--num_train_epochs="3" \
	--per_device_train_batch_size="4" \
	--gradient_accumulation_steps="1" \
	--learning_rate="3e-5" \
	--warmup_steps="500" \
	--evaluation_strategy="steps" \
	--audio_column_name="file" \
	--text_column_name="text" \
	--save_steps="400" \
	--eval_steps="100" \
	--logging_steps="1" \
	--layerdrop="0.0" \
	--save_total_limit="3" \
	--freeze_feature_extractor \
	--gradient_checkpointing \
	--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” \
	--fp16 \
	--group_by_length \
	--push_to_hub \
	--do_train --do_eval \