| model: |
| name: omniASR_CTC_300M_v2 |
| path: null |
| family: wav2vec2_asr |
| arch: null |
| config_overrides: null |
| dtype: float32 |
| mmap: false |
| compile: false |
| compile_options: |
| fullgraph: false |
| dynamic: false |
| mode: default |
| backend: inductor |
| backend_options: null |
| pretrained_encoder: |
| name: '' |
| path: null |
| family: wav2vec2_ssl |
| arch: null |
| config_overrides: null |
| dtype: float32 |
| mmap: false |
| compile: false |
| compile_options: |
| fullgraph: false |
| dynamic: false |
| mode: default |
| backend: inductor |
| backend_options: null |
| dataset: |
| name: example_dataset |
| family: null |
| config_overrides: |
| data: /asr-finetuning/fleurs-shona-omni-waxal-sna-omni-combined |
| train_split: train |
| valid_split: validation |
| storage_mode: MIXTURE_PARQUET |
| task_mode: ASR |
| manifest_storage_config: |
| sync_batches: true |
| sync_mode: UNTIL_FIRST |
| read_text: false |
| cached_fd_count: 1000 |
| mixture_parquet_storage_config: |
| sync_batches: true |
| sync_mode: UNTIL_FIRST |
| fragment_streaming: |
| parquet_path: '' |
| filesystem: null |
| name: null |
| weight: 1.0 |
| partition_filters: null |
| limit: |
| fraction_of_files: null |
| nb_files: null |
| nb_fragments: null |
| nb_rows: null |
| split_to_row_groups: true |
| seed: 2 |
| fragment_shuffle_window: 0 |
| files_circular_shift: false |
| nb_epochs: null |
| fragment_loading: |
| columns: {} |
| rename_columns: true |
| add_fragment_traces: false |
| drop_null: false |
| min_batch_size: 1 |
| filters: null |
| non_deterministic_read: false |
| use_threads: false |
| nb_prefetch: 1 |
| num_parallel_fragments: 1 |
| cache: true |
| cache_dir: null |
| dataset_summary_path: |
| /asr-finetuning/fleurs-shona-omni-waxal-sna-omni-combined/language_distribution_0.tsv |
| beta_corpus: 0.5 |
| beta_language: 0.0 |
| pa_cpu_count: 20 |
| max_workers: 30 |
| split_column: split |
| num_prefetch: 1 |
| parquet_path_name: _parquet_path |
| sampling_seed: 261 |
| asr_task_config: |
| min_audio_len: 32000 |
| max_audio_len: 960000 |
| batching_strategy: LENGTH |
| batch_size: 8 |
| num_seqs_multiple_of: 8 |
| max_num_elements: 960000 |
| max_bucket_size: null |
| drop_remainder: false |
| filter_long_text_threshold: null |
| remove_unknown: false |
| min_samples_per_char: 160 |
| normalize_audio: true |
| use_fbank: false |
| spec_aug_p: null |
| spec_aug_freq_mask_param: 80 |
| spec_aug_time_mask_param: 80 |
| unified_audio_feature_keys: true |
| example_shuffle_window: 1 |
| batch_shuffle_window: 1 |
| npc: 10 |
| seed: 2 |
| no_padding: false |
| max_num_batches: null |
| num_prefetch: 4 |
| tokenizer: |
| name: omniASR_tokenizer_v1 |
| path: null |
| family: null |
| config_overrides: null |
| gang: |
| tensor_parallel_size: 1 |
| timeout: 15 |
| high_priority: true |
| trainer: |
| data_parallelism: ddp |
| fsdp: |
| version: v2 |
| granularity: layer |
| hybrid: false |
| reshard_after_forward: true |
| fp32_reduce: true |
| mixed_precision: |
| mode: static |
| dtype: bfloat16 |
| grad_accumulation: |
| num_batches: 4 |
| no_sync: false |
| activation_checkpointing: |
| mode: off |
| every_nth_layer: 1 |
| max_grad_norm: null |
| fp16_loss_scale: |
| - 128.0 |
| - 0.0001 |
| gc_every_n_steps: 1000 |
| grad_check: false |
| anomaly_detection: false |
| freeze_encoder_for_n_steps: 0 |
| optimizer: |
| name: adamw |
| config: |
| lr: 3e-06 |
| betas: |
| - 0.9 |
| - 0.98 |
| eps: 1e-08 |
| weight_decay: 0.0 |
| amsgrad: false |
| maximize: false |
| capturable: false |
| differentiable: false |
| impl: auto |
| groups: [] |
| lr_scheduler: |
| name: tri_stage |
| config: |
| stage_ratio: |
| - 0.1 |
| - 0.4 |
| - 0.5 |
| start_lr_scale: 0.01 |
| final_lr_scale: 0.05 |
| regime: |
| num_steps: 5000 |
| num_data_epochs: null |
| validate_at_start: false |
| validate_after_n_steps: 2500 |
| validate_every_n_steps: 2500 |
| validate_after_n_data_epochs: 0 |
| validate_every_n_data_epochs: null |
| score_metric: wer |
| checkpoint_after_n_steps: 0 |
| checkpoint_every_n_steps: 2500 |
| checkpoint_after_n_data_epochs: 0 |
| checkpoint_every_n_data_epochs: null |
| save_model_only: false |
| export_hugging_face: false |
| keep_last_n_checkpoints: null |
| keep_best_n_checkpoints: null |
| keep_checkpoint_every_n_steps: null |
| publish_metrics_after_n_steps: 0 |
| publish_metrics_every_n_steps: 500 |
| publish_metrics_after_n_data_epochs: 0 |
| publish_metrics_every_n_data_epochs: null |
| common: |
| torch: |
| num_threads: null |
| allow_tf32: true |
| fp16_reduced_precision: true |
| bf16_reduced_precision: true |
| default_sdpa: torch |
| compiled_region_activation_memory_budget: 1.0 |
| metric_recorders: |
| tensorboard: |
| enabled: true |
| wandb: |
| enabled: false |
| entity: null |
| project: null |
| run_id: persistent |
| run_name: null |
| group: null |
| job_type: null |
| resume_mode: null |
| profilers: |
| torch: |
| enabled: false |
| skip_n_steps: 4 |
| wait_n_steps: 0 |
| num_warmup_steps: 1 |
| num_active_steps: 4 |
| repeat: 1 |
| assets: |
| extra_paths: [] |
| prev_checkpoint_dir: null |
| seed: 2 |
| debug: false |
| cluster: auto |
| no_sweep_dir: false |
| sweep_format: ws_{world_size}.{hash} |
|
|
