Saved in:
Bibliographic Details
Main Authors: Motwani, Sumeet Ramesh, Du, Chuan, Petrov, Aleksander, Davis, Christopher, Torr, Philip, Papania-Davis, Antonio, Yan, Weishi
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.16804
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918486287056896
author Motwani, Sumeet Ramesh
Du, Chuan
Petrov, Aleksander
Davis, Christopher
Torr, Philip
Papania-Davis, Antonio
Yan, Weishi
author_facet Motwani, Sumeet Ramesh
Du, Chuan
Petrov, Aleksander
Davis, Christopher
Torr, Philip
Papania-Davis, Antonio
Yan, Weishi
contents Optimization problems are central to decision-making in manufacturing, logistics, scheduling, and other industrial settings. Translating complicated descriptions of these problems into solver-ready formulations requires specialized operations research (OR) expertise, making it hard to scale. We present AutoOR, a scalable synthetic data generation and reinforcement learning pipeline that trains LLMs to autoformalize optimization problems specified in natural language across linear, mixed-integer, and non-linear categories. AutoOR generates verified training data from standard optimization forms and uses solver execution feedback as the reward signal for RL post-training. AutoOR applied to an 8B model achieves state-of-the-art or competitive results across six established OR benchmarks, matching significantly larger frontier models. For a non-linear problem class involving physical dynamics, where frontier models score near 0%, we introduce a curriculum RL strategy that bootstraps from limited initial training data to make this class tractable for post-training. We believe that methods such as AutoOR can significantly accelerate industrial decision-making with AI.
format Preprint
id arxiv_https___arxiv_org_abs_2604_16804
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems
Motwani, Sumeet Ramesh
Du, Chuan
Petrov, Aleksander
Davis, Christopher
Torr, Philip
Papania-Davis, Antonio
Yan, Weishi
Machine Learning
Artificial Intelligence
Optimization problems are central to decision-making in manufacturing, logistics, scheduling, and other industrial settings. Translating complicated descriptions of these problems into solver-ready formulations requires specialized operations research (OR) expertise, making it hard to scale. We present AutoOR, a scalable synthetic data generation and reinforcement learning pipeline that trains LLMs to autoformalize optimization problems specified in natural language across linear, mixed-integer, and non-linear categories. AutoOR generates verified training data from standard optimization forms and uses solver execution feedback as the reward signal for RL post-training. AutoOR applied to an 8B model achieves state-of-the-art or competitive results across six established OR benchmarks, matching significantly larger frontier models. For a non-linear problem class involving physical dynamics, where frontier models score near 0%, we introduce a curriculum RL strategy that bootstraps from limited initial training data to make this class tractable for post-training. We believe that methods such as AutoOR can significantly accelerate industrial decision-making with AI.
title AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2604.16804