hf token problem fix 2

util.py

@@ -210,33 +210,28 @@ class KaniModel:
         self.player = player
         self.hf_token = token
         self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
-        
+
         print(f"Loading model: {self.conf.model_name}")
         print(f"Target device: {self.device}")
-        
-        # Load model with proper configuration
-        load_kwargs = {
-            "dtype": torch.bfloat16,
-            "device_map": self.conf.device_map,
-            "trust_remote_code": True
-        }
+
+        # Set HF_TOKEN in environment to avoid parameter passing issues
         if self.hf_token:
-            load_kwargs["token"] = self.hf_token
+            os.environ['HF_TOKEN'] = self.hf_token
 
+        # Load model with proper configuration
+        # Don't pass token parameter - it will be read from HF_TOKEN env var
         self.model = AutoModelForCausalLM.from_pretrained(
             self.conf.model_name,
-            **load_kwargs
+            dtype=torch.bfloat16,
+            device_map=self.conf.device_map,
+            trust_remote_code=True
         )
 
-        tokenizer_kwargs = {"trust_remote_code": True}
-        if self.hf_token:
-            tokenizer_kwargs["token"] = self.hf_token
-
         self.tokenizer = AutoTokenizer.from_pretrained(
             self.conf.model_name,
-            **tokenizer_kwargs
+            trust_remote_code=True
         )
-        
+
         print(f"Model loaded successfully on device: {next(self.model.parameters()).device}")
 
     def get_input_ids(self, text_prompt: str, speaker_id:str) -> tuple[torch.tensor]: