| python run_flax_speech_recognition_seq2seq.py \ | |
| --dataset_name="google/xtreme_s" \ | |
| --model_name_or_path="sanchit-gandhi/flax-wav2vec2-2-bart-large-scan" \ | |
| --dataset_config_name="voxpopuli.en" \ | |
| --train_split_name="train" \ | |
| --eval_split_name="validation" \ | |
| --test_split_name="test" \ | |
| --dataset_cache_dir="/home/sanchitgandhi/cache/huggingface/datasets" \ | |
| --output_dir="./" \ | |
| --preprocessing_num_workers="1" \ | |
| --id_column_name="id" \ | |
| --text_column_name="transcription" \ | |
| --overwrite_output_dir \ | |
| --per_device_train_batch_size="8" \ | |
| --per_device_eval_batch_size="4" \ | |
| --logging_steps="25" \ | |
| --max_steps="50000" \ | |
| --eval_steps="10000" \ | |
| --save_steps="10000" \ | |
| --gradient_checkpointing \ | |
| --max_duration_in_seconds="20" \ | |
| --max_target_length="128" \ | |
| --generation_max_length="40" \ | |
| --generation_num_beams="1" \ | |
| --generation_length_penalty="1.2" \ | |
| --final_generation_max_length="200" \ | |
| --final_generation_num_beams="5" \ | |
| --learning_rate="1e-4" \ | |
| --warmup_steps="500" \ | |
| --save_total_limit="1" \ | |
| --freeze_feature_encoder \ | |
| --predict_with_generate \ | |
| --do_lower_case \ | |
| --do_eval \ | |
| --do_train \ | |
| --do_predict \ | |
| --push_to_hub \ | |
| --use_auth_token \ | |
| --wandb_project="voxpopuli" \ | |
| --wandb_name="flax-wav2vec2-2-bart-large-voxpopuli-baseline" | |