Push model using huggingface_hub.

best_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7023f9debdbd3e51ea534f4113bd982310de6912e5d106d2384f63f1a8f0282
+size 24544610

config.yaml

@@ -0,0 +1,109 @@
+datamodule:
+  _target_: look2hear.datas.datamodule.DataModule
+  batch_size: 1
+  num_workers: 8
+  pin_memory: true
+  DataClass:
+    _target_: look2hear.datas.datasets.waveform.WaveformDataClass
+    train_json_dir: /home/likai/ssd/Look2hear/examples/CTCNet/LRS2/tr
+    val_json_dir: /home/likai/ssd/Look2hear/examples/CTCNet/LRS2/cv
+    test_json_dir: /home/likai/ssd/Look2hear/examples/CTCNet/LRS2/tt
+    n_src: 1
+    task: enhancement
+    sample_rate: 16000
+    segment: 4
+    is_drop: false
+    normalize_audio: false
+    augmentation: false
+audiomodel:
+  _target_: look2hear.models.ctcnet.CTCNet
+  encoder_type: ConvolutionalEncoder
+  decoder_type: ConvolutionalDecoder
+  audio_channels: 1
+  audio_encoder_channels: 512
+  audio_encoder_kernels: 21
+  audio_encoder_strides: 10
+  audio_in_channels: 512
+  audio_out_channels: 512
+  audio_kernel_size: 5
+  audio_depth: 4
+  audio_block_type: ConvNormAct
+  audio_norm_type: gLN
+  audio_act_type: PReLU
+  audio_shared: true
+  visual_encoder_channels: 512
+  visual_in_channels: 64
+  visual_out_channels: 64
+  visual_kernel_size: 3
+  visual_depth: 4
+  visual_block_type: ConvNormAct
+  visual_norm_type: BatchNorm1d
+  visual_act_type: PReLU
+  visual_shared: false
+  fusion_type: ConcatFusion
+  fusion_shared: false
+  n_repeats: 3
+  m_repeats: 13
+  mask_types: MaskGenerator
+  num_speakers: 1
+  mask_kernel_size: 1
+  mask_act: ReLU
+  mask_RI_split: false
+  mask_output_gate: false
+  mask_dw_gate: false
+  mask_direct: false
+  mask_is2d: false
+videomodel:
+  _target_: look2hear.video_models.resnetmodel.ResNetVideoModel
+  activation_type: PReLU
+  pretrained: /home/likai/ssd/Look2hear/pretrain_zoo/frcnn_128_512.backbone.pth.tar
+audio_optimizer:
+  _target_: torch.optim.AdamW
+  lr: 0.001
+  weight_decay: 0.1
+audio_scheduler:
+  _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
+  mode: min
+  factor: 0.5
+  patience: 10
+audio_loss:
+  _target_: look2hear.losses.pitwrapper.PITLossWrapper
+  loss_func: look2hear.losses.snr.neg_sisdr
+  pit: true
+  mode: permutation-wise
+  eval_func: min
+system:
+  _target_: look2hear.systems.single_speaker.SingleSpeaker
+  freeze_video_model: true
+  compile: false
+exp:
+  dir: /home/likai/ssd/Look2hear/examples/CTCNet
+  name: CTCNet-1
+checkpoint:
+  _target_: lightning.pytorch.callbacks.ModelCheckpoint
+  dirpath: ${exp.dir}/${exp.name}/checkpoints
+  monitor: val/neg_sisdr
+  mode: min
+  verbose: true
+  save_top_k: 1
+  save_last: true
+  filename: '{epoch}-{val/neg_sisdr:.4f}'
+logger:
+  _target_: lightning.pytorch.loggers.WandbLogger
+  name: ${exp.name}
+  save_dir: ${exp.dir}/${exp.name}/logs
+  offline: true
+  project: Look2hear
+trainer:
+  _target_: lightning.pytorch.Trainer
+  devices:
+  - 0
+  max_epochs: 5
+  sync_batchnorm: true
+  gradient_clip_val: 5.0
+  default_root_dir: ${exp.dir}/${exp.name}/
+  accelerator: cuda
+  limit_train_batches: 0.001
+  limit_val_batches: 0.001
+  fast_dev_run: false
+  precision: bf16-mixed