A physics-aware two-stage SFT+GRPO training method with period/AST/sandbox rewards raises small open LLMs from ~0-2% to ~68-77% on a strict OpenSeesPy building-modeling benchmark (BMEval).
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Rethinking Scientific Modeling: Toward Physically Consistent and Simulation-Executable Programmatic Generation
A physics-aware two-stage SFT+GRPO training method with period/AST/sandbox rewards raises small open LLMs from ~0-2% to ~68-77% on a strict OpenSeesPy building-modeling benchmark (BMEval).