trust_remote_code=True

utils/models.py

@@ -101,8 +101,10 @@ def run_inference(model_name, context, question):
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     result = ""
-    model_kwargs = {} # make sure qwen3 doesn't use thinking
-    if "qwen3" in model_name.lower(): # Making it case-insensitive and checking for substring
+    model_kwargs = {
+        "trust_remote_code": True,
+    } # make sure qwen3 doesn't use thinking
+    if "qwen3" in model_name.lower(): 
         print(f"Recognized {model_name} as a Qwen3 model. Setting enable_thinking=False.")
         model_kwargs["enable_thinking"] = False