This supervised fine-tuning (SFT) dataset is built from the 2018 Howard University Physics Graduate Qualifying Exam, pairing questions across classical and modern physics with full step-by-step solutions.
Potential Use Cases
- Supervised Fine-Tuning for Scientific Reasoning: Trains LLMs to construct multi-step derivations across both classical and modern physics, reinforcing framework selection, approximation handling, and limit checking as explicit reasoning steps.
- Broad-Coverage Physics Evaluation: Serves as an evaluation set for general physics competence, revealing where a model's reasoning holds in classical regimes but degrades in relativistic and quantum settings.