| command: | |
| - python3 | |
| - ${program} | |
| - --overwrite_output_dir | |
| - --freeze_feature_encoder | |
| - --gradient_checkpointing | |
| - --predict_with_generate | |
| - --fp16 | |
| - --group_by_length | |
| - --do_train | |
| - --do_eval | |
| - --load_best_model_at_end | |
| - --push_to_hub | |
| - --use_auth_token | |
| - ${args} | |
| method: random | |
| metric: | |
| goal: maximize | |
| name: eval/bleu | |
| parameters: | |
| model_name_or_path: | |
| value: ./ | |
| task: | |
| value: covost2 | |
| language: | |
| value: fr.en | |
| eval_split_name: | |
| value: test | |
| output_dir: | |
| value: ./output_dir | |
| num_train_epochs: | |
| value: 3 | |
| per_device_train_batch_size: | |
| value: 4 | |
| per_device_eval_batch_size: | |
| value: 4 | |
| gradient_accumulation_steps: | |
| value: 8 | |
| generation_max_length: | |
| value: 40 | |
| generation_num_beams: | |
| value: 1 | |
| learning_rate: | |
| distribution: log_uniform | |
| max: -6.9 | |
| min: -9.2 | |
| hidden_dropout: | |
| distribution: log_uniform | |
| max: -1.6 | |
| min: -3.4 | |
| warmup_steps: | |
| value: 500 | |
| evaluation_strategy: | |
| value: steps | |
| max_duration_in_seconds: | |
| value: 20 | |
| save_steps: | |
| value: 500 | |
| eval_steps: | |
| value: 500 | |
| logging_steps: | |
| value: 1 | |
| metric_for_best_model: | |
| value: bleu | |
| greater_is_better: | |
| value: True | |
| program: run_xtreme_s.py | |
| project: xtreme_s_xlsr_2_bart_covost2_fr_en | |