anyone knows how to limit thinking length in vllm openai server? it seems like need to adit the logits processor.