Microsoft Research has just released OptiMind, “a small language model designed to convert business problems described in natural language into the mathematical formulations needed by optimization software. Built on a 20-billion parameter model, OptiMind is compact by today’s standards yet matches the performance of larger, more complex systems. Its modest size means it can run locally on users’ devices, enabling fast iteration while keeping sensitive business data on users’ devices rather than transmitting it to external servers.“

OptiMind incorporates knowledge from optimization experts both during training and when it’s being used to improve formulation accuracy at scale. Three stages enable this: domain-specific hints improve training data quality, the model undergoes fine-tuning, and expert reasoning guides the model as it works.

MS warns that OptiMind is an experimental model to gather community feedback and inform future development. Link