| CUDA_VISIBLE_DEVICES=0 python run_xtreme_s.py \ | |
| --model_name_or_path="./" \ | |
| --task="covost2" \ | |
| --language="fr.en" \ | |
| --eval_split_name="test" \ | |
| --output_dir="./" \ | |
| --overwrite_output_dir \ | |
| --num_train_epochs="10" \ | |
| --per_device_train_batch_size="8" \ | |
| --per_device_eval_batch_size="8" \ | |
| --gradient_accumulation_steps="4" \ | |
| --generation_max_length="40" \ | |
| --generation_num_beams="1" \ | |
| --learning_rate="3e-4" \ | |
| --warmup_steps="500" \ | |
| --evaluation_strategy="steps" \ | |
| --max_duration_in_seconds="20" \ | |
| --save_steps="500" \ | |
| --eval_steps="500" \ | |
| --logging_steps="1" \ | |
| --freeze_feature_encoder \ | |
| --gradient_checkpointing \ | |
| --fp16 \ | |
| --group_by_length \ | |
| --do_train \ | |
| --do_eval \ | |
| --metric_for_best_model="bleu" \ | |
| --greater_is_better="True" \ | |
| --load_best_model_at_end \ | |
| --push_to_hub \ | |
| --predict_with_generate \ | |
| --use_auth_token | |