fix: Remove torch.cuda.is_available() to prevent CUDA init in main process

app.py

@@ -149,8 +149,7 @@ def upload_model_as_pr(
 @spaces.GPU(duration=30)
 def completion_signal():
     """Brief GPU allocation at completion to satisfy ZeroGPU requirements."""
-    import torch
-    return torch.cuda.is_available()
+    return True
 
 
 def run_quantization(config):