2026-07-08 08:57:50 +00:00

5.3 KiB
Raw Permalink Blame History

Humanity's-Last-Exam

概述

Humanity's Last ExamHLE是一个综合性语言模型基准测试包含2,500道涵盖广泛学科的问题。该基准由人工智能安全中心Center for AI Safety与Scale AI联合创建是目前最具挑战性的学术基准之一。

任务描述

  • 任务类型:专家级问答
  • 输入问题14%为多模态,含图片)
  • 输出:答案、解释及置信度分数
  • 领域分布数学41%、物理9%)、生物/医学11%)、计算机科学/AI10%、人文9%、工程4%、化学7%、其他9%

主要特点

  • 跨多个学科的2,500道专家级问题
  • 14%的问题需要多模态理解能力
  • 24%为选择题76%为简答精确匹配题
  • 问题来源涵盖各类学术与专业领域
  • 响应格式包含置信度评分

评估说明

  • 默认使用 test 数据划分进行评估
  • 主要指标基于大语言模型LLM裁判的 准确率Accuracy
  • 响应格式包括解释Explanation、答案Answer和置信度Confidence0100%
  • 注意:对于纯文本模型,请将 extra_params["include_multi_modal"] 设为 False
  • 使用 GRADE: C/I 格式进行 LLM 裁判评分

属性

属性
基准测试名称 hle
数据集ID cais/hle
论文 N/A
标签 Knowledge, QA
指标 acc
默认示例数Shots 0-shot
评估划分 test

数据统计

指标
总样本数 2,500
提示词长度(平均) 1029.85 字符
提示词长度(最小/最大) 234 / 21341 字符

各子集统计信息:

子集 样本数 提示平均长度 提示最小长度 提示最大长度
Biology/Medicine 280 1259.39 246 13702
Chemistry 165 812.72 236 6942
Computer Science/AI 241 1581.02 263 11529
Engineering 111 1620.26 250 21341
Humanities/Social Science 219 1069.39 256 7028
Math 1,021 862.46 262 8952
Physics 230 1027.63 257 17139
Other 233 754.94 234 13655

图像统计信息:

指标
图像总数 342
每样本图像数 最小: 1, 最大: 1, 平均: 1
分辨率范围 329x12 14950x2780
图像格式 gif, jpeg, png, webp

样例示例

子集: Biology/Medicine

{
  "input": [
    {
      "id": "906a518f",
      "content": "Your response should be in the following format:\nExplanation: {your explanation for your answer choice}\nAnswer: {your chosen answer}\nConfidence: {your confidence score between 0% and 100% for your answer}"
    },
    {
      "id": "d03d8d4e",
      "content": [
        {
          "text": "In a bioinformatics lab, Watterson's estimator (theta) and pi (nucleotide diversity) will be calculated from variant call files which contain human phased samples with only single nucleotide variants present, and there are no completely missi ... [TRUNCATED] ... y pi (nucleotide diversity) is biased.\nC. Both Watterson's estimator (theta) and pi (nucleotide diversity) are biased.\nD. Neither Watterson's estimator (theta) nor pi (nucleotide diversity) are biased.\nE. None of the other answers are correct"
        }
      ]
    }
  ],
  "target": "B",
  "id": 0,
  "group_id": 0,
  "subset_key": "Biology/Medicine",
  "metadata": {
    "uid": "66e88728ba7d8bc0d5806f3a",
    "author_name": "Scott S",
    "rationale": "First, we recognize that all single nucleotide variants are included somewhere in the sample. It is given that, across “all samples,” there are no “missing single nucleotide variants.” Further, since “[t]he number of samples is arbitrarily la ... [TRUNCATED] ... fferent genotypes that that position, the analysis would consider these two genomes to have the same nucleotide at the position. This reduces the estimated nucleotide diversity, pi. Therefore, pi would be biased in the circumstance described.",
    "raw_subject": "Bioinformatics",
    "category": "Biology/Medicine",
    "has_image": false
  }
}

注:部分内容因展示需要已被截断。

提示模板

提示模板:

{question}

额外参数

参数 类型 默认值 描述
include_multi_modal bool True 评估时是否包含多模态(图像)问题。

使用方法

通过命令行CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets hle \
    --limit 10  # 正式评估时请删除此行

通过 Python

from evalscope import run_task
from evalscope.config import TaskConfig

task_cfg = TaskConfig(
    model='YOUR_MODEL',
    api_url='OPENAI_API_COMPAT_URL',
    api_key='EMPTY_TOKEN',
    datasets=['hle'],
    dataset_args={
        'hle': {
            # subset_list: ['Biology/Medicine', 'Chemistry', 'Computer Science/AI']  # 可选,用于评估特定子集
            # extra_params: {}  # 使用默认额外参数
        }
    },
    limit=10,  # 正式评估时请删除此行
)

run_task(task_cfg=task_cfg)