python run_pseudo_labelling.py \ | |
--model_name_or_path "openai/whisper-large-v2" \ | |
--dataset_name "sanchit-gandhi/librispeech_asr_clean" \ | |
--dataset_config_name "clean" \ | |
--data_split_name "train.100" \ | |
--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" \ | |
--wandb_name "whisper-large-v2-beam-libri-train.clean.100" \ | |
--per_device_eval_batch_size 16 \ | |
--max_label_length 256 \ | |
--dtype "bfloat16" \ | |
--preprocessing_num_workers 16 \ | |
--report_to "wandb" \ | |
--dataloader_num_workers 16 \ | |
--streaming False \ | |
--generation_num_beams 1 | |