remove explicit tokenizer device assignment

utils/models.py

@@ -240,7 +240,7 @@ def run_inference(model_name, context, question):
                 return_tensors="pt",
                 return_dict=True,
                 **tokenizer_kwargs,
-            ).to(device)
+            ).to(bitnet_model.device)
             with torch.inference_mode():
                 # Check interrupt before generation
                 if generation_interrupt.is_set():