debug

app.py

@@ -38,7 +38,7 @@ tokenizer = AutoTokenizer.from_pretrained(LLM_MODEL_PATH)
 mllm = Qwen2_5_VLForConditionalGeneration.from_pretrained(MLLM_MODEL_PATH, torch_dtype=torch.bfloat16, device_map="auto")
 llm = Qwen2ForCausalLM.from_pretrained(LLM_MODEL_PATH, torch_dtype=torch.bfloat16, device_map="auto")
 
-mllm_sampling = dict(temperature=0, max_new_tokens=8192)
+mllm_sampling = dict(do_sample=False, temperature=0, max_new_tokens=8192)
 llm_sampling = dict(temperature=0.6, top_p=0.95, max_new_tokens=8192)
 
 # === Build Prompts ===