| buffer_size: 5000 | |
| checkpointing_steps: 500 | |
| dataloader_num_workers: 10 | |
| dataloader_pin_memory: false | |
| dataset_in_streaming_mode: false | |
| dataset_name: dataset | |
| debug_mode: false | |
| delta_t: 10 | |
| do_lower_case: false | |
| do_remove_punctuation: false | |
| evaluation_steps: 500 | |
| gradient_accumulation_steps: 4 | |
| hub_model_id: null | |
| hub_token: null | |
| init_r: 12 | |
| language: Bulgarian | |
| language_abbr: bg | |
| learning_rate: 5.0e-05 | |
| load_best_model: false | |
| logging_steps: 100 | |
| lora_alpha: 32 | |
| lora_dropout: 0.1 | |
| lr_scheduler_type: linear | |
| max_audio_input_length: 30.0 | |
| max_train_steps: 5600 | |
| model_name_or_path: openai/whisper-large-v2 | |
| num_train_epochs: 20 | |
| num_warmup_steps: 0 | |
| orth_reg_weight: 0.5 | |
| output_dir: IliyanGochev/whisper-large-v2-bg-peft | |
| overwrite_cache: false | |
| per_device_eval_batch_size: 32 | |
| per_device_train_batch_size: 8 | |
| preprocessing_num_workers: 20 | |
| push_to_hub: true | |
| r: 8 | |
| report_to: tensorboard | |
| resume_from_checkpoint: null | |
| seed: null | |
| target_r: 4 | |
| task: transcribe | |
| tfinal: 1000 | |
| tinit: 200 | |
| use_adalora: true | |
| use_peft: true | |
| weight_decay: 0.0 | |
| with_tracking: true | |