instruction-following benchmark for large reasoning models - View it on GitHub