Sys2Bench
收藏资源简介:
Sys2Bench是一个全面评估大型语言模型在推理和规划能力上的基准,包含11个数据集,涵盖算术推理、逻辑推理、常识推理、算法推理和规划五大类别。这些数据集的任务包括解决算术问题、逻辑问题、利用常识进行推理、解决算法问题以及进行规划。Sys2Bench的构建旨在评估不同推理和规划任务中现有推理时间技术的效果,以推动大型语言模型在这方面的能力提升。
Sys2Bench is a benchmark for comprehensively evaluating the reasoning and planning capabilities of large language models. It consists of 11 datasets covering five major categories: arithmetic reasoning, logical reasoning, commonsense reasoning, algorithmic reasoning, and planning. The tasks included in these datasets cover solving arithmetic problems, logical problems, reasoning via commonsense, solving algorithmic problems, and carrying out planning. Sys2Bench is constructed to assess the effectiveness of existing reasoning-time techniques across diverse reasoning and planning tasks, so as to promote the improvement of large language models' capabilities in this field.

- 1Inference-Time Computations for LLM Reasoning and Planning: A Benchmark and Insights德克萨斯A&M大学计算机科学与工程系 · 2025年



