LLMsPark
收藏资源简介:
LLMsPark是一个基于博弈论的评价平台,用于衡量大型语言模型(LLMs)在经典博弈论环境中的决策策略和社会行为。该平台提供了多智能体环境,以探索战略深度。LLMsPark使用排行榜排名和评分机制对15个主流LLMs进行交叉评估,包括商业和开源模型。LLMsPark旨在丰富现有的评估基准,并扩展LLMs在交互式、博弈论场景中的评估。
LLMsPark is a game theory-based evaluation platform designed to measure the decision-making strategies and social behaviors of large language models (LLMs) in classic game-theoretic scenarios. This platform offers multi-agent environments for exploring strategic depth. LLMsPark implements a leaderboard ranking and scoring framework to conduct cross-evaluation across 15 mainstream LLMs, including both commercial and open-source models. The platform aims to enrich existing evaluation benchmarks and expand the assessment of LLMs in interactive, game-theoretic scenarios.
LLMsParks 数据集概述
数据集基本信息
- 数据集名称: LLMsParks
- 数据更新日期: 2023年10月20日
数据集内容
数据集包含多个模型在不同游戏中的性能评估结果。
评估游戏列表
- Who Is Spy
- The Prisoners Dilemma
- The Trust Game
- The Nim Game
- The Dictator Game
评估维度
每个游戏均包含以下评估维度:
- Multiple Rounds
- Single Round
模型性能数据
| 模型名称 | 参数量 | Who Is Spy | The Prisoners Dilemma | The Trust Game | The Nim Game | The Dictator Game |
|---|---|---|---|---|---|---|
| Baichuan-7B | 7B | - | 877.36 | 1020.57 | 952.37 | 1106.68 |
| phoenix-inst-chat-7b | 7B | - | 858.29 | 1236.50 | 973.40 | 910.56 |
| chatglm-6b | 6B | - | 863.98 | 895.12 | 872.35 | 850.90 |
| chatglm2-6b | 6B | - | 880.60 | 1210.51 | 926.13 | 1285.09 |
| ChatYuan-large-v2 | - | - | 1165.18 | 866.28 | 917.13 | 913.76 |
| moss-moon-003-sft | 16B | 73 | 906.92 | 912.06 | 937.68 | 923.80 |
| dolly-v2-12b | 12B | - | 906.54 | 899.30 | 1179.13 | 858.29 |
| ChatGLM-Pro | - | 60 | 1046.46 | 1234.31 | 968.27 | 1335.80 |
| CharacterGLM | - | - | 1073.42 | 1230.59 | 953.36 | 1343.28 |
| gpt-3.5-turbo | - | 60 | 1174.41 | 851.80 | 1161.53 | 912.35 |
| gpt-4 | - | 59 | 1285.80 | 1062.07 | 1247.80 | 871.80 |
| RWKV-4-World-7B | 7B | - | 922.54 | 923.56 | 950.85 | 887.63 |
| Baichuan2-7B-Chat | 7B | - | 905.10 | 799.67 | 882.21 | 888.08 |
| minimax-abab5-chat | - | 62 | 1133.40 | 857.60 | 1077.79 | 911.91 |
| Qwen-14B-Chat | 14B | 77 | - | - | - | - |




