Hi, i'm curious that I want to use VLM model to extract accident documents so a few of accident vocab don't exist in LLM vocab. What should I do ?