Fine-tuning large language models for an interpretability challenge (compute costs) | grantmaking.ai