| python run_speech_recognition_ctc.py \ | |
| --dataset_name="mozilla-foundation/common_voice_8_0" \ | |
| --dataset_config_name="ka" \ | |
| --model_name_or_path="facebook/wav2vec2-xls-r-1b" \ | |
| --output_dir="./" \ | |
| --overwrite_output_dir \ | |
| --max_steps 1000 \ | |
| --per_device_train_batch_size="16" \ | |
| --per_device_eval_batch_size="64" \ | |
| --gradient_accumulation_steps="8" \ | |
| --dataloader_num_workers 8 \ | |
| --learning_rate="8e-5" \ | |
| --lr_scheduler_type cosine \ | |
| --adam_beta2 0.98 \ | |
| --warmup_ratio 0.1 \ | |
| --evaluation_strategy="steps" \ | |
| --text_column_name="sentence" \ | |
| --chars_to_ignore \, \? \. \! \- \; \: \" \“ \% \‘ ” „ \� \' « » \( \) ։ ՝ ՞ ՛ ՚ \ | |
| --save_steps="100" \ | |
| --eval_steps="100" \ | |
| --logging_steps="100" \ | |
| --save_total_limit="2" \ | |
| --freeze_feature_encoder \ | |
| --layerdrop="0.1" \ | |
| --activation_dropout="0.1" \ | |
| --feat_proj_dropout="0.0" \ | |
| --mask_time_prob="0.75" \ | |
| --mask_time_length="10" \ | |
| --mask_feature_prob="0.25" \ | |
| --mask_feature_length="64" \ | |
| --gradient_checkpointing \ | |
| --use_auth_token \ | |
| --fp16 \ | |
| --group_by_length \ | |
| --do_train --do_eval \ | |
| --load_best_model_at_end \ | |
| --report_to all \ | |
| --run_name="xlsr-ka-cv-1b-1" \ | |
| --wandb_project="xlsr-ka" \ | |
| --bnb --tristage_sched | |