Merge pull request #191 from magic-akari/4.1-Stable

feat: add compress_model

README.md

@@ -293,6 +293,15 @@ The existing steps before clustering do not need to be changed. All you need to
 
 ### [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/svc-develop-team/so-vits-svc/blob/4.1-Stable/sovits4_for_colab.ipynb) [sovits4_for_colab.ipynb](https://colab.research.google.com/github/svc-develop-team/so-vits-svc/blob/4.1-Stable/sovits4_for_colab.ipynb)
 
+## 🗜️ Model strip
+
+The generated model contains data that is needed for further training. If you confirm that the model is final and not be used in further training, it is safe to strip these data to get smaller file size (about 1/3).
+
+```shell
+# Example
+python compress_model.py -c="configs/config.json" -i="logs/44k/G_30400.pth" -o="logs/44k/release.pth"
+```
+
 ## 📤 Exporting to Onnx
 
 Use [onnx_export.py](https://github.com/svc-develop-team/so-vits-svc/blob/4.0/onnx_export.py)

README_zh_CN.md

@@ -295,6 +295,17 @@ python inference_main.py -m "logs/44k/G_30400.pth" -c "configs/config.json" -n "
 
 ### [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/svc-develop-team/so-vits-svc/blob/4.1-Stable/sovits4_for_colab.ipynb) [sovits4_for_colab.ipynb](https://colab.research.google.com/github/svc-develop-team/so-vits-svc/blob/4.1-Stable/sovits4_for_colab.ipynb)
 
+## 🗜️ 模型压缩
+
+生成的模型含有继续训练所需的信息。如果确认不再训练，可以移除模型中此部分信息，得到约 1/3 大小的最终模型。
+
+使用 [compress_model.py](compress_model.py)
+
+```shell
+# 例
+python compress_model.py -c="configs/config.json" -i="logs/44k/G_30400.pth" -o="logs/44k/release.pth"
+```
+
 ## 📤 Onnx导出
 
 使用 [onnx_export.py](onnx_export.py)

compress_model.py

@@ -0,0 +1,69 @@
+from collections import OrderedDict
+
+import torch
+
+import utils
+from models import SynthesizerTrn
+
+
+def copyStateDict(state_dict):
+    if list(state_dict.keys())[0].startswith('module'):
+        start_idx = 1
+    else:
+        start_idx = 0
+    new_state_dict = OrderedDict()
+    for k, v in state_dict.items():
+        name = ','.join(k.split('.')[start_idx:])
+        new_state_dict[name] = v
+    return new_state_dict
+
+
+def removeOptimizer(config: str, input_model: str, output_model: str):
+    hps = utils.get_hparams_from_file(config)
+
+    net_g = SynthesizerTrn(hps.data.filter_length // 2 + 1,
+                           hps.train.segment_size // hps.data.hop_length,
+                           **hps.model)
+
+    optim_g = torch.optim.AdamW(net_g.parameters(),
+                                hps.train.learning_rate,
+                                betas=hps.train.betas,
+                                eps=hps.train.eps)
+
+    state_dict_g = torch.load(input_model, map_location="cpu")
+    new_dict_g = copyStateDict(state_dict_g)
+    keys = []
+    for k, v in new_dict_g['model'].items():
+        keys.append(k)
+
+    new_dict_g = {k: new_dict_g['model'][k] for k in keys}
+
+    torch.save(
+        {
+            'model': new_dict_g,
+            'iteration': 0,
+            'optimizer': optim_g.state_dict(),
+            'learning_rate': 0.0001
+        }, output_model)
+
+
+if __name__ == "__main__":
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-c",
+                        "--config",
+                        type=str,
+                        default='configs/config.json')
+    parser.add_argument("-i", "--input", type=str)
+    parser.add_argument("-o", "--output", type=str, default=None)
+
+    args = parser.parse_args()
+
+    output = args.output
+
+    if output is None:
+        import os.path
+        filename, ext = os.path.splitext(args.input)
+        output = filename + "_release" + ext
+
+    removeOptimizer(args.config, args.input, output)