run.sh

#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES="1" python run_speech_recognition_ctc.py \
        --dataset_name="timit_asr" \
        --model_name_or_path="microsoft/unispeech-large-1500h-cv" \
        --overwrite_output_dir \
        --output_dir="./unispeech-large-1500h-cv-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 \