python run_speech_recognition_ctc.py \ | |
--dataset_name="common_voice" \ | |
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \ | |
--dataset_config_name="tr" \ | |
--output_dir="./" \ | |
--overwrite_output_dir \ | |
--num_train_epochs="100" \ | |
--per_device_train_batch_size="64" \ | |
--gradient_accumulation_steps="1" \ | |
--learning_rate="5e-4" \ | |
--warmup_steps="500" \ | |
--evaluation_strategy="steps" \ | |
--text_column_name="sentence" \ | |
--save_steps="500" \ | |
--eval_steps="500" \ | |
--logging_steps="1" \ | |
--layerdrop="0.0" \ | |
--eval_metrics wer cer \ | |
--save_total_limit="1" \ | |
--mask_time_prob="0.3" \ | |
--mask_time_length="10" \ | |
--mask_feature_prob="0.1" \ | |
--mask_feature_length="64" \ | |
--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \ | |
--fp16 \ | |
--group_by_length \ | |
--push_to_hub \ | |
--do_train --do_eval \ | |
--gradient_checkpointing | |