Update app.py

app.py

@@ -1681,6 +1681,10 @@ class Hive:
             self.pipe = _remote_pipe
             # For remote inference, we still need a local tokenizer for prompt compilation
             self.tok = AutoTokenizer.from_pretrained(self.model_id, trust_remote_code=trust, chat_template=None)
+            # We pass `token=False` to prevent from_pretrained from using a potentially invalid
+            # environment token, as we only need the public tokenizer config.
+            # The actual inference call uses the token provided to the InferenceClient.
+            self.tok = AutoTokenizer.from_pretrained(self.model_id, trust_remote_code=trust, chat_template=None, token=False)
             self.model = None # No local model needed
             self.stopping_criteria = None # Not used with InferenceClient
         else: