Update app.py

app.py

@@ -504,7 +504,7 @@ def load_llama_model(model_path, enable_thinking):
                 mul_mat_q=True,
                  use_mmap=True,
                 chat_handler=Qwen35ChatHandler(
-      clip_model_path=f"models/mmproj-BF16.gguf",
+      clip_model_path=f"models/mmproj-F32.gguf",
       # force_reasoning=True,
                     enable_thinking=enable_thinking ,
       image_min_tokens=1024, # Note: Qwen-VL models require at minimum 1024 image tokens to function correctly on bbox grounding tasks