if it's a fine-tuned model then pass... that should work as a custom llm