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="pa-IN" \
        --output_dir="./" \
        --overwrite_output_dir \
        --num_train_epochs="200" \
        --per_device_train_batch_size="16" \
        --per_device_eval_batch_size="32" \
        --learning_rate="1.11e-4" \
        --gradient_accumulation_steps="2" \
        --warmup_steps="2000" \
        --length_column_name="input_length" \
        --evaluation_strategy="steps" \
        --text_column_name="sentence" \
        --save_steps="500" \
        --eval_steps="500" \
        --logging_steps="500" \
        --layerdrop="0.02" \
        --attention_dropout="0.05" \
        --activation_dropout="0.05" \
        --hidden_dropout="0.04" \
        --feat_proj_dropout="0.01" \
        --layerdrop="0.02" \
        --ctc_loss_reduction="mean" \
        --save_total_limit="1" \
        --freeze_feature_encoder \
        --mask_time_prob="0.20" \
        --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – « » „ \` _ \
        --gradient_checkpointing \
        --use_auth_token \
        --fp16 \
        --group_by_length \
        --do_train --do_eval \
        --push_to_hub