run.sh

python run_speech_recognition_ctc.py \
    --dataset_name="mozilla-foundation/common_voice_8_0" \
    --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
    --dataset_config_name="ab" \
    --output_dir="./" \
    --overwrite_output_dir \
    --num_train_epochs="100" \
    --per_device_train_batch_size="16" \
    --per_device_eval_batch_size="8" \
    --gradient_accumulation_steps="2" \
    --learning_rate="7.5e-5" \
    --warmup_steps="2000" \
    --length_column_name="input_length" \
    --evaluation_strategy="steps" \
    --text_column_name="sentence" \
    --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � \' \’ \– \
    --save_steps="300" \
    --eval_steps="300" \
    --logging_steps="300" \
    --layerdrop="0.025" \
    --activation_dropout="0.085" \
    --save_total_limit="2" \
    --freeze_feature_encoder \
    --feat_proj_dropout="0.04" \
    --mask_time_prob="0.30" \
    --mask_time_length="10" \
    --mask_feature_prob="0.40" \
    --mask_feature_length="64" \
    --gradient_checkpointing \
    --use_auth_token \
    --fp16 \
    --group_by_length \
    --do_train --do_eval \
    --push_to_hub