Tg Develop's W Okada Fork Cloud

The web interface (client) is just a control panel; the actual voice conversion and backend processes happen on the server. If the cloud server crashes or fails to launch properly, you can enable Debug Mode to read the exact error logs directly in your notebook's output cell.

To do this, you need to append the --log-level debug argument to the command that launches the server.

!./ML_Program --log-level debug

!./VoiceChanger --log-level debug

The notebook output cell will now print detailed debug logs. If the server crashes, copy the text output from that cell. Save it to a .txt file, or paste it to a site like Pastebin to share with others when asking for support.

• This utility allows you to convert model weights both ways: PyTorch (.pth) to Safetensors (.safetensors), and Safetensors (.safetensors) back to PyTorch (.pth).
• Safetensors ➜ PyTorch: W-Okada saves imported models internally as .safetensors files inside your model_dir folder. If you accidentally deleted your original .pth files, you can use this script to recover them.
• PyTorch ➜ Safetensors: You can also convert standard .pth weights to .safetensors for safer sharing, faster loading, or local archiving.

Step 1: Install Requirements

Step 2: Create the Script

• Create a file named safetensors-pytorch-converter.py and paste the code below into it.

• (For Google Colab & Kaggle, you can automatically create this file by running a new code cell with %%writefile safetensors-pytorch-converter.py as the very first line, followed by the script below).

  #  RVC Model Weight Converter (PTH <-> Safetensors)
  
  
  import argparse
  import os
  import sys
  import torch
  from safetensors.torch import load_file, save_file
  
  def convert_pth_to_safetensors(pth_path, output_path):
      # Loads a PyTorch .pth model and exports its state_dict to Safetensors
      print(f"Reading PyTorch weights from: {pth_path}")
      try:
          # weights_only=True prevents execution of arbitrary code during load
          weights = torch.load(pth_path, map_location="cpu", weights_only=True)
  
          # Unpack the state_dict if the file contains training metadata
          if isinstance(weights, dict) and "state_dict" in weights:
              weights = weights["state_dict"]
  
          print(f"Writing Safetensors weights to: {output_path}")
          save_file(weights, output_path)
          print("Conversion successful.")
      except Exception as e:
          print(f"Error during PTH conversion: {e}", file=sys.stderr)
  
  
  def convert_safetensors_to_pth(safetensors_path, output_path):
      # Loads a Safetensors model and exports it as a PyTorch .pth file
      print(f"Reading Safetensors weights from: {safetensors_path}")
      try:
          weights = load_file(safetensors_path, device="cpu")
          print(f"Writing PyTorch weights to: {output_path}")
          torch.save(weights, output_path)
          print("Conversion successful.")
      except Exception as e:
          print(f"Error during Safetensors conversion: {e}", file=sys.stderr)
  
  
  def main():
      parser = argparse.ArgumentParser(
          description="Convert RVC weights between PyTorch (.pth) and Safetensors formats."
      )
      parser.add_argument(
          "input_file", 
          type=str, 
          help="Path to the .pth or .safetensors file to convert"
      )
      parser.add_argument(
          "-o", "--output", 
          type=str, 
          default=None, 
          help="Optional custom output path"
      )
  
      args = parser.parse_args()
      input_path = args.input_file
  
      if not os.path.exists(input_path):
          print(f"Error: The file '{input_path}' does not exist.", file=sys.stderr)
          sys.exit(1)
  
      file_name, file_ext = os.path.splitext(input_path)
      ext = file_ext.lower()
  
      if ext == ".pth":
          output_path = args.output if args.output else f"{file_name}.safetensors"
          convert_pth_to_safetensors(input_path, output_path)
      elif ext == ".safetensors":
          output_path = args.output if args.output else f"{file_name}.pth"
          convert_safetensors_to_pth(input_path, output_path)
      else:
          print("Error: Input file must be a .pth or .safetensors file.", file=sys.stderr)
          sys.exit(1)
  
  
  if __name__ == "__main__":
      main()

Step 3: Run the Script