Fix remaining device_map auto in gradio_app.py

gradio_app.py

@@ -56,7 +56,7 @@ class ModelManager:
             self.model = AutoModelForCausalLM.from_pretrained(
                 base_model_name,
                 torch_dtype=torch.float16 if self.device == "cuda:0" else torch.float32,
-                device_map="auto" if self.device == "cuda:0" else None,
+                device_map={"": 0}  # Force all parameters to GPU 0,
                 trust_remote_code=True,
                 low_cpu_mem_usage=True,
                 token=hf_token