billwang37/optim-ai-7b-v1

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Apr 23, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

OptimAI 7B v1 by billwang37 is a 7.6 billion parameter language model, fine-tuned from Qwen2.5-Math-7B, specifically designed for operations research and optimal control problems. It excels at mathematically formulating, solving, and explaining concepts across various optimization domains. This model is specialized for tasks like linear programming, network flow, inventory problems, and LQR/Riccati optimal control, providing explicit bridges between OR and control theory.

Loading preview...

Overview

OptimAI 7B v1 is a 7.6 billion parameter model developed by Bill Wang, fine-tuned from Qwen2.5-Math-7B. It specializes in operations research and optimal control, aiming to formulate problems mathematically, solve them, and explain the theoretical connections between these fields.

Key Capabilities

  • Problem Formulation & Solving: Handles natural-language descriptions to formulate and solve optimization problems.
  • Operations Research: Proficient in linear and integer programming (e.g., diet, knapsack), network flow (max flow, shortest path), inventory problems (newsvendor, EOQ), and basic queuing (M/M/c).
  • Optimal Control: Supports LQR/Riccati optimal control and explains KKT conditions and LP duality.
  • Theoretical Explanations: Connects optimization theory with optimal control concepts.

Training Details

The model was fine-tuned using LoRA (r=64, alpha=128) on approximately 1,600 synthetic OR/control problems, with a 4-bit base (bitsandbytes NF4). It also incorporated DPO training with around 200 preference pairs. The LoRA adapter has been merged into the base weights for this release.

Limitations

  • Primarily trained on synthetic data, which may lead to incorrect formulations for out-of-distribution problems.
  • Performance is strong on listed problem classes but experimental for areas like stochastic programming, robust optimization, and PDE-constrained optimization.
  • Intended as a research demonstration; solutions should always be verified with dedicated solvers.

Good For

  • Developers and researchers working on operations research and optimal control applications.
  • Educational purposes to understand problem formulation and theoretical connections.
  • Rapid prototyping of solutions for well-defined optimization problems.

Finetunes
1 models