update

src/models/Llama.py

@@ -24,15 +24,15 @@ class Llama(Model):
         ]
 
     def _load_model_if_needed(self):
-        if self._model is None:
+        if self.model is None:
             model = AutoModelForCausalLM.from_pretrained(
                 self.name,
                 torch_dtype=torch.bfloat16,
                 token=self.hf_token,
                 device_map="auto",  # or omit entirely to default to CPU
             )
-            self._model = model
-        return self._model
+            self.model = model
+        return self.model
 
     def query(self, msg, max_tokens=128000):
         model = self._load_model_if_needed().to("cuda")