| base_config: ${base_config_path}/train_ddp.yaml | |
| sr: 16000 | |
| n_fft: 512 | |
| hop_length: 160 | |
| n_mic: 8 | |
| n_src: 6 | |
| dim_latent: 64 | |
| autocast: True | |
| trainer: | |
| max_epochs: 200 | |
| gradient_clip_val: 5.0 | |
| sync_batchnorm: True | |
| precision: 16-mixed | |
| benchmark: True | |
| use_distributed_sampler: False | |
| callbacks: | |
| - _target_: lightning.pytorch.callbacks.ModelCheckpoint | |
| filename: "{epoch:04d}" | |
| save_last: True | |
| save_top_k: -1 | |
| - _target_: lightning.pytorch.callbacks.RichProgressBar | |
| refresh_rate: 5 | |
| - _target_: lightning.pytorch.callbacks.RichModelSummary | |
| max_depth: 3 | |
| - _target_: neural_fcasa.callbacks.cyclic_annealer.CyclicAnnealerCallback | |
| name: beta | |
| cycle: 10 | |
| max_value: 1.0 | |
| ini_period: 50 | |
| ini_max_value: 1.0 | |
| - _target_: neural_fcasa.callbacks.visualizer.VisualizerCallback | |
| dataset_name: chunk.derev-hop${hop_length} | |
| datamodule: | |
| _target_: neural_fcasa.datamodules.hdf5_wavact_datamodule.DataModule | |
| train_dataset_path: ${working_directory}/../../hdf5/${dataset_name}-tr.hdf5 | |
| val_dataset_path: ${working_directory}/../../hdf5/${dataset_name}-cv.hdf5 | |
| batch_size: 2 | |
| duration: 10 | |
| sr: ${sr} | |
| hop_length: ${hop_length} | |
| task: | |
| _target_: neural_fcasa.tasks.avi_scl_allsort_task.AVITask | |
| n_fft: ${n_fft} | |
| hop_length: ${hop_length} | |
| n_src: ${n_src} | |
| beta: 1.0 | |
| gamma: 1.0 | |
| encoder: | |
| _target_: neural_fcasa.encoders.resepformer_encoder.RESepFormerEncoder | |
| n_fft: ${n_fft} | |
| n_mic: ${n_mic} | |
| n_src: ${n_src} | |
| dim_latent: ${dim_latent} | |
| chunk_size: 100 | |
| step_size: 100 | |
| d_model: 256 | |
| dim_feedforward: 1024 | |
| n_blocks: 8 | |
| norm_first: True | |
| autocast: ${autocast} | |
| diagonalizer: | |
| _target_: neural_fcasa.diagonalizers.iss_nrmxt_diagonalizer.ISSDiagonalizer | |
| eps: 1.e-4 | |
| n_iter: 2 | |
| spec_aug: | |
| _target_: torchaudio.transforms.SpecAugment | |
| n_time_masks: 2 | |
| time_mask_param: 64 | |
| n_freq_masks: 2 | |
| freq_mask_param: 20 | |
| zero_masking: True | |
| decoder: | |
| _target_: neural_fcasa.decoders.res_lin_decoder.Decoder | |
| n_fft: ${n_fft} | |
| dim_latent: ${dim_latent} | |
| dim_latent_noi: 10 | |
| optimizer_config: | |
| _target_: aiaccel.torch.lightning.OptimizerConfig | |
| optimizer_generator: | |
| _partial_: True | |
| _target_: torch.optim.AdamW | |
| lr: 1.e-4 | |
| weight_decay: 1.e-5 | |