Writing an LLM from scratch, part 26 – evaluating the fine-tuned model