Update huggingface_llm.py

huggingface_llm.py

@@ -16,13 +16,13 @@ class HuggingFaceLLM(LLM):
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
-        if self.device == "cpu":
-            self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        self.device = "cuda" if torch.cuda.is_available() and self.device != "cpu" else "cpu"
         self._load_model()
 
     def _load_model(self):
         self._tokenizer = AutoTokenizer.from_pretrained(self.model_id)
-        self._model = AutoModelForCausalLM.from_pretrained(self.model_id).to(self.device)
+        self._model = AutoModelForCausalLM.from_pretrained(self.model_id)
+        self._model = self._model.to(torch.device(self.device))
 
     @property
     def _llm_type(self) -> str: