Optimize GPU device handling in generate function

- Simplify model device transfer method
- Remove redundant `.to("cuda")` assignment
- Ensure model is correctly moved to GPU when available

webui.py

@@ -59,7 +59,7 @@ def generate(text,
     
     # if gpu available, move model to gpu
     if torch.cuda.is_available():
-        model = model.to("cuda")
+        model.to("cuda")
     
     with torch.no_grad():
         wav = model.inference(