HeuriGym
收藏资源简介:
HeuriGym是一个为评估大型语言模型(LLMs)在组合优化问题中生成的启发式算法而设计的代理框架。数据集包含九个组合优化问题,旨在评估LLMs的多步推理能力。数据集由康奈尔大学创建,旨在解决现有评估方法在评估LLMs的推理和基于代理的问题解决能力方面的不足。HeuriGym使用明确的指标和迭代细化来评估LLMs的性能,并提供了一个开放源代码的基准套件,旨在推动LLMs在科学和工程领域更有效和现实的问题解决能力的发展。
HeuriGym is a proxy framework designed for evaluating heuristic algorithms generated by large language models (LLMs) for combinatorial optimization problems. The dataset comprises nine combinatorial optimization tasks, intended to assess the multi-step reasoning capabilities of LLMs. Developed by Cornell University, this dataset was created to address the deficiencies of existing evaluation methods in assessing LLMs’ reasoning and agent-based problem-solving abilities. HeuriGym employs explicit metrics and iterative refinement to evaluate LLM performance, and provides an open-source benchmark suite aimed at promoting more effective and practical problem-solving capabilities of LLMs in the fields of science and engineering.

- 1HeuriGym: An Agentic Benchmark for LLM-Crafted Heuristics in Combinatorial Optimization康奈尔大学 · 2025年



