update weights loader

hifigan/utils.py

@@ -3,7 +3,7 @@
 import sys
 import torch
 import torchaudio
-from typing import List, Tuple
+from typing import Any, Dict, List, Tuple
 
 MATPLOTLIB_FLAG = False
 
@@ -55,4 +55,21 @@ def summarize(writer, global_step, scalars={}, histograms={}, images={}, audios=
     for k, v in images.items():
         writer.add_image(k, v, global_step, dataformats='HWC')
     for k, v in audios.items():
-        writer.add_audio(k, v, global_step, audio_sampling_rate)
+        writer.add_audio(k, v, global_step, audio_sampling_rate)
+
+
+def load_state_dict(model_state_dict, state_dict: Dict[str, Any]) -> None:
+    is_changed = False
+    for k in state_dict:
+        if k in model_state_dict:
+            if state_dict[k].shape != model_state_dict[k].shape:
+                logging.info(f"Skip loading parameter: {k}, "
+                            f"required shape: {model_state_dict[k].shape}, "
+                            f"loaded shape: {state_dict[k].shape}")
+                # state_dict[k] = model_state_dict[k]
+                del state_dict[k]
+                is_changed = True
+        else:
+            logging.info(f"Dropping parameter {k}")
+            is_changed = True
+    return state_dict

save_state.py

@@ -2,6 +2,14 @@
 import glob
 import torch
 from hifigan.model.hifigan import HifiGAN
+
+def save(ckpt_path: str):
+    model = HifiGAN.load_from_checkpoint(checkpoint_path=ckpt_path, strict=True)
+    # print(model.net_g.state_dict())
+    torch.save(model.net_g.state_dict(), "net_g.pt")
+    torch.save(model.net_period_d.state_dict(), "net_period_d.pt")
+    torch.save(model.net_scale_d.state_dict(), "net_scale_d.pt")
+
 def main():
     ckpt_path = None
     if os.path.exists("logs/lightning_logs"):
@@ -13,10 +21,6 @@ def main():
                 ckpt_path = last_ckpt
 
     print(ckpt_path)
-
-    model = HifiGAN.load_from_checkpoint(checkpoint_path=ckpt_path, strict=False)
-
-    torch.save(model.net_g.state_dict(), "out.pt")
-
+    save("logs/lightning_logs_v1/version_8/checkpoints/last.ckpt")
 if __name__ == "__main__":
     main()

train.py

@@ -20,6 +20,7 @@
 
 from hifigan.hparams import HParams
 from hifigan.data.dataset import MelDataset, MelDataset
+from hifigan.utils import load_state_dict
 
 def get_hparams(config_path: str) -> HParams:
     with open(config_path, "r") as f:
@@ -47,7 +48,10 @@ def main():
     valid_loader = DataLoader(valid_dataset, batch_size=1, num_workers=16, shuffle=False, pin_memory=True, collate_fn=collate_fn)
 
     model = HifiGAN(**hparams)
-    # model = HifiGAN.load_from_checkpoint(checkpoint_path="logs/lightning_logs/version_10/checkpoints/last.ckpt", strict=False)
+    # model = HifiGAN.load_from_checkpoint(checkpoint_path="logs/lightning_logs_v1/version_8/checkpoints/last.ckpt", strict=True)
+    # model.net_g._load_from_state_dict(torch.load("net_g.pt"), strict=False)
+    # model.net_period_d._load_from_state_dict(torch.load("net_period_d.pt"), strict=False)
+    # model.net_scale_d._load_from_state_dict(torch.load("net_scale_d.pt"), strict=False)
 
     checkpoint_callback = ModelCheckpoint(dirpath=None, save_last=True, every_n_train_steps=2000, save_weights_only=False)