Update app.py

app.py

@@ -336,7 +336,7 @@ def ensure_model_loaded():
     attn_impl = "eager"
     # Use GPU if available, otherwise CPU
     if torch.cuda.is_available():
-        dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+        dtype = torch.bfloat16  # Use bfloat16 on GPU for consistency
         device_map = "auto"
     else:
         dtype = torch.float32