python run_pseudo_labelling.py \ | |
--model_name_or_path "openai/whisper-tiny" \ | |
--dataset_name "distil-whisper/librispeech_asr" \ | |
--dataset_config_name "all" \ | |
--data_split_name "validation.clean+validation.other" \ | |
--text_column_name "text" \ | |
--cache_dir "/home/sanchitgandhi/.cache" \ | |
--dataset_cache_dir "/home/sanchitgandhi/.cache" \ | |
--output_dir "./transcriptions-streaming" \ | |
--wandb_dir "/home/sanchitgandhi/.cache" \ | |
--wandb_project "distil-whisper-debug" \ | |
--per_device_eval_batch_size 1 \ | |
--dtype "bfloat16" \ | |
--dataloader_num_workers 16 \ | |
--logging_steps 2 \ | |
--report_to "wandb" \ | |
--streaming \ | |
--max_samples_per_split 256 \ | |
--max_label_length 256 \ | |
--return_timestamps \ | |
--decode_token_ids False | |