We convert all the LLM provider to OpenAI format and this preprocessing can add up 10-20ms but apart from it. I don't think you should see any additional latency.