Results for fine-tuning open-weight LLMs on dedicated bare metal GPUs