| model: | |
| _target_: src.model.ConformerModel | |
| input_dim: 128 | |
| writer: | |
| _target_: src.logger.CometMLWriter | |
| project_name: pytorch_template_asr_example | |
| workspace: null | |
| run_name: conformer_30m | |
| mode: online | |
| loss_names: | |
| - loss | |
| log_checkpoints: false | |
| id_length: 32 | |
| run_id: m2guzao93o9ytjxogwt78mftkyiqalsf | |
| metrics: | |
| train: [] | |
| inference: | |
| - _target_: src.metrics.ArgmaxCERMetric | |
| name: CER_(Argmax) | |
| - _target_: src.metrics.ArgmaxWERMetric | |
| name: WER_(Argmax) | |
| - _target_: src.metrics.WER | |
| name: WER | |
| - _target_: src.metrics.CER | |
| name: CER | |
| datasets: | |
| train: | |
| _target_: src.datasets.LibrispeechDataset | |
| part: train-other-500 | |
| instance_transforms: ${transforms.instance_transforms.train} | |
| val: | |
| _target_: src.datasets.LibrispeechDataset | |
| part: test-clean | |
| instance_transforms: ${transforms.instance_transforms.inference} | |
| test: | |
| _target_: src.datasets.LibrispeechDataset | |
| part: test-other | |
| instance_transforms: ${transforms.instance_transforms.inference} | |
| dataloader: | |
| _target_: torch.utils.data.DataLoader | |
| batch_size: 30 | |
| num_workers: 2 | |
| pin_memory: true | |
| transforms: | |
| instance_transforms: | |
| train: | |
| get_spectrogram: | |
| _target_: torchaudio.transforms.MelSpectrogram | |
| sample_rate: 16000 | |
| audio: | |
| _target_: torchvision.transforms.v2.Compose | |
| transforms: | |
| - _target_: src.transforms.wav_augs.Gain | |
| sample_rate: 16000 | |
| min_gain_in_db: -6 | |
| max_gain_in_db: 6 | |
| p: 0.2 | |
| - _target_: src.transforms.wav_augs.Shift | |
| p: 0.2 | |
| - _target_: src.transforms.wav_augs.PitchShift | |
| min_semitones: -2 | |
| max_semitones: 2 | |
| p: 0.2 | |
| - _target_: src.transforms.wav_augs.Noise | |
| p: 0.3 | |
| inference: | |
| get_spectrogram: | |
| _target_: torchaudio.transforms.MelSpectrogram | |
| sample_rate: 16000 | |
| batch_transforms: | |
| train: null | |
| inference: null | |
| optimizer: | |
| _target_: torch.optim.AdamW | |
| lr: 5.0e-05 | |
| lr_scheduler: | |
| _target_: torch.optim.lr_scheduler.OneCycleLR | |
| max_lr: 0.0001 | |
| pct_start: 0.1 | |
| steps_per_epoch: ${trainer.epoch_len} | |
| epochs: ${trainer.n_epochs} | |
| anneal_strategy: cos | |
| loss_function: | |
| _target_: src.loss.CTCLossWrapper | |
| text_encoder: | |
| _target_: src.text_encoder.CTCTextEncoder | |
| trainer: | |
| log_step: 200 | |
| n_epochs: 150 | |
| epoch_len: 1300 | |
| device_tensors: | |
| - spectrogram | |
| - text_encoded | |
| resume_from: checkpoint-epoch62.pth | |
| device: auto | |
| override: false | |
| monitor: min val_WER_(Argmax) | |
| save_period: 5 | |
| early_stop: ${trainer.n_epochs} | |
| save_dir: saved | |
| seed: 1 | |