Researchers have developed a new framework called SDDL that improves the accuracy of language models in solving combinatorial optimization problems. These problems involve finding feasible solutions within large search spaces while satisfying complex constraints. In resource-constrained settings, smaller language models often struggle to preserve feasibility when scheduling directly from natural language. SDDL translates natural-language scheduling problems into compact representations that are more easily solved by external solvers. This approach allows even the smallest models to achieve high levels of accuracy in solving these problems. --- Why it matters: This research matters because it enables resource-constrained language models to tackle complex combinatorial optimization tasks, which is crucial for applications such as natural language processing and artificial intelligence. By improving the feasibility and optimality of solutions, SDDL has significant implications for the development of more efficient and effective AI systems. Source: https://arxiv.org/abs/2608.18409