Installation Fails

What happened?

Installing and Launching this app fails without modifying the code.

Steps to reproduce

1. Install VibeVoice within Pinokio
2. Installation script turns bright blue and fails.

Your system (OS / GPU / RAM / VRAM / etc.)
Windows 11, 32Gb of RAM with NVidia 4070 16Gb with Cuda

Logs / full error output

[startup] Found 25 voice presets
   [startup] Loading voice preset en-Carter_man from
   C:\pinokio\api\vibevoice-realtime.git\app\demo\voices\streaming_model\en-Car 
   ter_man.pt
   [startup] Loading prefilled prompt from
   C:\pinokio\api\vibevoice-realtime.git\app\demo\voices\streaming_model\en-Car 
   ter_man.pt
   ERROR:    Traceback (most recent call last):
     File
   "C:\pinokio\api\vibevoice-realtime.git\app\env\lib\site-packages\starlette\r 
   outing.py", line 638, in lifespan
       async with self.lifespan_context(app) as maybe_state:
     File
   "C:\pinokio\api\vibevoice-realtime.git\app\env\lib\site-packages\fastapi\rou 
   ting.py", line 241, in __aenter__
       await self._router._startup()
     File
   "C:\pinokio\api\vibevoice-realtime.git\app\env\lib\site-packages\fastapi\rou 
   ting.py", line 4884, in _startup
       await handler()
     File "C:\pinokio\api\vibevoice-realtime.git\app\demo\web\app.py", line     
   353, in _startup
       service.load()
     File "C:\pinokio\api\vibevoice-realtime.git\app\demo\web\app.py", line     
   126, in load
       self._ensure_voice_cached(self.default_voice_key)
     File "C:\pinokio\api\vibevoice-realtime.git\app\demo\web\app.py", line     
   164, in _ensure_voice_cached
       prefilled_outputs = torch.load(
     File
   "C:\pinokio\api\vibevoice-realtime.git\app\env\lib\site-packages\torch\seria 
   lization.py", line 1578, in load
       raise pickle.UnpicklingError(_get_wo_message(str(e))) from None
   _pickle.UnpicklingError: Weights only load failed. In PyTorch 2.6, we        
   changed the default value of the `weights_only` argument in `torch.load`     
   from `False` to `True`. Re-running `torch.load` with `weights_only` set to   
   `False` will likely succeed, but it can result in arbitrary code execution.  
   Do it only if you got the file from a trusted source.
   Please file an issue with the following so that we can make
   `weights_only=True` compatible with your use case: WeightsUnpickler error:   
                                                                                
   Can only SETITEMS for dict, collections.OrderedDict, collections.Counter,    
   but got <class 'transformers.modeling_outputs.BaseModelOutputWithPast'>      
                                                                                
   Check the documentation of torch.load to learn more about types accepted by  
   default with weights_only
   https://pytorch.org/docs/stable/generated/torch.load.html.
                                                                                
   ERROR:    Application startup failed. Exiting.
                                                                                
   (env) (base) C:\pinokio\api\vibevoice-realtime.git\app>

Recommended Changes:

Re-applying Fixes: 
  Re-applying and strictly verifying the PyTorch compatibility fixes.

  ✓  Edit  app.py 

      164       prefilled_outputs = torch.load(
      165           preset_path,
      166           map_location=self._torch_device,
      167 -         weights_only=True,
      167 +         weights_only=False,
      168       )
      169   self._voice_cache[key] = prefilled_outputs
      170


  ✓  Edit  realtime_model_inference_from_file.py 

      225   voice_sample = voice_mapper.get_voice_path(args.speaker_name)       
      226  print(f"Using voice preset for {args.speaker_name}:
           {voice_sample}")
      227   with torch.serialization.safe_globals([BaseModelOutputWithPast,     
            DynamicCache]):
      228 -     all_prefilled_outputs = torch.load(voice_sample,
          map_location=target_device, weights_only=True)
      228 +     all_prefilled_outputs = torch.load(voice_sample,
          map_location=target_device, weights_only=False)
      229
      230   # Prepare inputs for the model
      231   inputs = processor.process_input_with_cached_prompt(

  ✓  Edit  vibevoice_tokenizer_processor.py 

      222       return audio_array
      223   elif file_ext == '.pt':
      224       # PyTorch tensor file
      225 -     audio_tensor = torch.load(audio_path, map_location='cpu',       
          weights_only=True).squeeze()
      225 +     audio_tensor = torch.load(audio_path, map_location='cpu',       
          weights_only=False).squeeze()
      226       if isinstance(audio_tensor, torch.Tensor):
      227           audio_array = audio_tensor.numpy()
      228       else: