Evaluation Report — qwen3-max

Dataset	Model	Metric	Score	Num
GPQA-Diamond	qwen3-max	mean_acc	0.9000	10
GSM8K	qwen3-max	mean_acc	1.0000	10
IFEval	qwen3-max	mean_prompt_level_strict	0.8000	10
IQuiz	qwen3-max	mean_acc	0.8000	20

GPQA-Diamond

0.9000

Description

Overview

GPQA (Graduate-Level Google-Proof Q&A) Diamond is a challenging benchmark of 198 multiple-choice questions written by domain experts in biology, physics, and chemistry. The questions are designed to be extremely difficult, requiring PhD-level expertise to answer correctly.

Task Description

Task Type: Expert-Level Multiple-Choice Q&A
Input: Graduate-level science question with 4 choices
Output: Single correct answer letter (A, B, C, or D)
Domains: Biology, Physics, Chemistry

Key Features

198 questions written and validated by domain PhD experts
Questions are "Google-proof" - cannot be easily looked up
Designed to test deep domain knowledge and reasoning
Diamond subset represents the highest quality questions
Average human expert accuracy ~65%, non-expert ~34%

Evaluation Notes

Default configuration uses 0-shot or 5-shot evaluation
Supports Chain-of-Thought (CoT) prompting for improved reasoning
Answer choices are randomly shuffled during evaluation
Only uses train split (validation set is private)
Challenging benchmark for measuring expert-level reasoning

Properties

Property	Value
Benchmark Name	`gpqa_diamond`
Dataset ID	AI-ModelScope/gpqa_diamond
Paper	N/A
Tags	`Knowledge`, `MCQ`
Metrics	`acc`
Default Shots	0-shot
Evaluation Split	`train`

Data Statistics

Metric	Value
Total Samples	198
Prompt Length (Mean)	841.15 chars
Prompt Length (Min/Max)	340 / 5845 chars

Sample Example

Subset: default

{
  "input": [
    {
      "id": "82b448a9",
      "content": "Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D. Think step by step before answering.\n\nTwo quantum states wi ... [TRUNCATED] ...  and 10^-8 sec, respectively. We want to clearly distinguish these two energy levels. Which one of the following options could be their energy difference so that they can be clearly resolved?\n\n\nA) 10^-4 eV\nB) 10^-9 eV\nC) 10^-8 eV\nD) 10^-11 eV"
    }
  ],
  "choices": [
    "10^-4 eV",
    "10^-9 eV",
    "10^-8 eV",
    "10^-11 eV"
  ],
  "target": "A",
  "id": 0,
  "group_id": 0,
  "subset_key": "",
  "metadata": {
    "correct_answer": "10^-4 eV",
    "incorrect_answers": [
      "10^-11 eV",
      "10^-8 eV\n",
      "10^-9 eV"
    ]
  }
}

Note: Some content was truncated for display.

Prompt Template

Prompt Template:

Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.

{question}

{choices}

概述

GPQA（Graduate-Level Google-Proof Q&A）Diamond 是一个极具挑战性的基准测试，包含 198 道由生物学、物理学和化学领域的专家编写的多项选择题。这些问题设计得极为困难，需要博士级别的专业知识才能正确作答。

任务描述

任务类型：专家级多项选择问答
输入：研究生水平的科学问题，附带 4 个选项
输出：单个正确答案字母（A、B、C 或 D）
领域：生物学、物理学、化学

主要特点

198 道题目均由相关领域的博士专家编写并验证
题目“无法通过 Google 轻易查到”——难以通过简单搜索获得答案
旨在测试深层次的领域知识与推理能力
Diamond 子集代表了最高质量的问题
人类专家平均准确率约为 65%，非专家约为 34%

评估说明

默认配置使用 0-shot 或 5-shot 评估
支持思维链（Chain-of-Thought, CoT）提示以提升推理能力
评估过程中答案选项会随机打乱
仅使用训练集（验证集为私有）
是衡量专家级推理能力的高难度基准测试

属性

属性	值
基准测试名称	`gpqa_diamond`
数据集 ID	AI-ModelScope/gpqa_diamond
论文	N/A
标签	`Knowledge`, `MCQ`
指标	`acc`
默认示例数（Shots）	0-shot
评估划分	`train`

数据统计

指标	值
总样本数	198
提示词长度（平均）	841.15 字符
提示词长度（最小/最大）	340 / 5845 字符

样例示例

子集: default

{
  "input": [
    {
      "id": "82b448a9",
      "content": "Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of A,B,C,D. Think step by step before answering.\n\nTwo quantum states wi ... [TRUNCATED] ...  and 10^-8 sec, respectively. We want to clearly distinguish these two energy levels. Which one of the following options could be their energy difference so that they can be clearly resolved?\n\n\nA) 10^-4 eV\nB) 10^-9 eV\nC) 10^-8 eV\nD) 10^-11 eV"
    }
  ],
  "choices": [
    "10^-4 eV",
    "10^-9 eV",
    "10^-8 eV",
    "10^-11 eV"
  ],
  "target": "A",
  "id": 0,
  "group_id": 0,
  "subset_key": "",
  "metadata": {
    "correct_answer": "10^-4 eV",
    "incorrect_answers": [
      "10^-11 eV",
      "10^-8 eV\n",
      "10^-9 eV"
    ]
  }
}

注：部分内容因展示需要已被截断。

提示模板

提示模板：

Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: [LETTER]' (without quotes) where [LETTER] is one of {letters}. Think step by step before answering.

{question}

{choices}

Category	Subset	Metric	Score	Num
default	default	mean_acc	0.9000	10

GPQA-Diamond · Subset Scores

GSM8K

1.0000

Description

Overview

GSM8K (Grade School Math 8K) is a high-quality dataset of 8.5K linguistically diverse grade school math word problems created by human problem writers. The dataset is specifically designed to evaluate and improve the multi-step mathematical reasoning capabilities of language models.

Task Description

Task Type: Mathematical Word Problem Solving
Input: Natural language math word problem
Output: Numerical answer derived through step-by-step reasoning
Difficulty: Grade school level (2-8 reasoning steps required)

Key Features

Problems require basic arithmetic operations (addition, subtraction, multiplication, division)
Solutions involve 2 to 8 sequential reasoning steps
High linguistic diversity in problem formulations
Human-written problems ensuring natural language quality
Clear numerical answers for objective evaluation

Evaluation Notes

Default configuration uses 4-shot examples with Chain-of-Thought (CoT) prompting
Answers should be formatted within \boxed{} for proper extraction
The metric extracts numerical values for accuracy comparison
Supports both zero-shot and few-shot evaluation modes

Properties

Property	Value
Benchmark Name	`gsm8k`
Dataset ID	AI-ModelScope/gsm8k
Paper	Paper
Tags	`Math`, `Reasoning`
Metrics	`acc`
Default Shots	4-shot
Evaluation Split	`test`
Train Split	`train`

Data Statistics

Metric	Value
Total Samples	1,319
Prompt Length (Mean)	1966.87 chars
Prompt Length (Min/Max)	1800 / 2575 chars

Sample Example

Subset: main

{
  "input": [
    {
      "id": "0bc7f97b",
      "content": "Here are some examples of how to solve similar problems:\n\nNatalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?\n\nReasoning:\nNatalia sold 48/ ... [TRUNCATED] ... ds every day with four. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. How much in dollars does she make every day at the farmers' market?\nPlease reason step by step, and put your final answer within \\boxed{}."
    }
  ],
  "target": "18",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "reasoning": "Janet sells 16 - 3 - 4 = <<16-3-4=9>>9 duck eggs a day.\nShe makes 9 * 2 = $<<9*2=18>>18 every day at the farmer’s market."
  }
}

Note: Some content was truncated for display.

Prompt Template

Prompt Template:

{question}
Please reason step by step, and put your final answer within \boxed{{}}.

Few-shot Template

Here are some examples of how to solve similar problems:

{fewshot}

{question}
Please reason step by step, and put your final answer within \boxed{{}}.

概述

GSM8K（Grade School Math 8K）是一个高质量的数据集，包含由人工编写者创作的 8.5K 道语言多样化的中小学数学应用题。该数据集专门用于评估和提升语言模型的多步数学推理能力。

任务描述

任务类型：数学应用题求解
输入：自然语言描述的数学应用题
输出：通过逐步推理得出的数值答案
难度：小学水平（需 2–8 步推理）

主要特点

问题涉及基本算术运算（加、减、乘、除）
解答过程包含 2 到 8 个连续推理步骤
问题表述具有高度的语言多样性
人工编写确保自然语言质量
答案为明确的数值，便于客观评估

评估说明

默认配置使用 4-shot 示例并结合思维链（Chain-of-Thought, CoT）提示
答案应使用 \boxed{} 格式包裹以便正确提取
评估指标会提取数值进行准确率比较
支持零样本（zero-shot）和少样本（few-shot）两种评估模式

属性

属性	值
基准测试名称	`gsm8k`
数据集ID	AI-ModelScope/gsm8k
论文	Paper
标签	`Math`, `Reasoning`
指标	`acc`
默认样本数	4-shot
评估划分	`test`
训练划分	`train`

数据统计

指标	值
总样本数	1,319
提示词长度（平均）	1966.87 字符
提示词长度（最小/最大）	1800 / 2575 字符

样例示例

子集: main

{
  "input": [
    {
      "id": "0bc7f97b",
      "content": "Here are some examples of how to solve similar problems:\n\nNatalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?\n\nReasoning:\nNatalia sold 48/ ... [TRUNCATED] ... ds every day with four. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. How much in dollars does she make every day at the farmers' market?\nPlease reason step by step, and put your final answer within \\boxed{}."
    }
  ],
  "target": "18",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "reasoning": "Janet sells 16 - 3 - 4 = <<16-3-4=9>>9 duck eggs a day.\nShe makes 9 * 2 = $<<9*2=18>>18 every day at the farmer’s market."
  }
}

注：部分内容因展示需要已被截断。

提示模板

提示模板：

{question}
Please reason step by step, and put your final answer within \boxed{{}}.

少样本模板

Here are some examples of how to solve similar problems:

{fewshot}

{question}
Please reason step by step, and put your final answer within \boxed{{}}.

Category	Subset	Metric	Score	Num
default	main	mean_acc	1.0000	10

GSM8K · Subset Scores

IFEval

0.8000

Description

Overview

IFEval (Instruction-Following Eval) is a benchmark for evaluating how well language models follow explicit, verifiable instructions. It contains prompts with specific formatting, content, or structural requirements that can be objectively verified.

Task Description

Task Type: Instruction Following Evaluation
Input: Prompts with explicit, verifiable constraints
Output: Response that follows all specified instructions
Constraint Types: Format, length, keywords, structure, etc.

Key Features

~500 prompts with 25 types of verifiable instructions
Instructions are objectively checkable (not subjective)
Examples: "write exactly 3 paragraphs", "include the word X", "use bullet points"
Tests instruction comprehension and compliance
No ambiguity in evaluation criteria

Evaluation Notes

Default configuration uses 0-shot evaluation
Four metrics available:
prompt_level_strict: All instructions in prompt must be followed
prompt_level_loose: Some tolerance for minor deviations
inst_level_strict: Per-instruction accuracy (strict)
inst_level_loose: Per-instruction accuracy (loose)
prompt_level_strict is the primary metric
Automatic verification of instruction compliance

Properties

Property	Value
Benchmark Name	`ifeval`
Dataset ID	opencompass/ifeval
Paper	N/A
Tags	`InstructionFollowing`
Metrics	`prompt_level_strict`, `inst_level_strict`, `prompt_level_loose`, `inst_level_loose`
Default Shots	0-shot
Evaluation Split	`train`

Data Statistics

Metric	Value
Total Samples	541
Prompt Length (Mean)	210.75 chars
Prompt Length (Min/Max)	53 / 1858 chars

Sample Example

Subset: default

{
  "input": [
    {
      "id": "cb71907f",
      "content": "Write a 300+ word summary of the wikipedia page \"https://en.wikipedia.org/wiki/Raymond_III,_Count_of_Tripoli\". Do not use any commas and highlight at least 3 sections that has titles in markdown format, for example *highlighted section part 1*, *highlighted section part 2*, *highlighted section part 3*."
    }
  ],
  "target": "",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "key": 1000,
    "prompt": "Write a 300+ word summary of the wikipedia page \"https://en.wikipedia.org/wiki/Raymond_III,_Count_of_Tripoli\". Do not use any commas and highlight at least 3 sections that has titles in markdown format, for example *highlighted section part 1*, *highlighted section part 2*, *highlighted section part 3*.",
    "instruction_id_list": [
      "punctuation:no_comma",
      "detectable_format:number_highlighted_sections",
      "length_constraints:number_words"
    ],
    "kwargs": [
      {
        "num_highlights": null,
        "relation": null,
        "num_words": null,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      },
      {
        "num_highlights": 3,
        "relation": null,
        "num_words": null,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      },
      {
        "num_highlights": null,
        "relation": "at least",
        "num_words": 300,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      }
    ]
  }
}

Prompt Template

No prompt template defined.

概述

IFEval（Instruction-Following Eval）是一个用于评估语言模型遵循明确、可验证指令能力的基准测试。它包含一系列带有特定格式、内容或结构要求的提示（prompts），这些要求可以被客观地验证。

任务描述

任务类型：指令遵循评估
输入：包含明确、可验证约束条件的提示
输出：完全遵循所有指定指令的响应
约束类型：格式、长度、关键词、结构等

主要特点

包含约500个提示，涵盖25种可验证的指令类型
指令可被客观检查（非主观判断）
示例：“写恰好3段文字”、“包含单词X”、“使用项目符号列表”
测试模型对指令的理解与遵守能力
评估标准无歧义

评估说明

默认配置采用 0-shot 评估方式
提供四种评估指标：
prompt_level_strict：提示中的所有指令都必须被严格遵循
prompt_level_loose：允许对轻微偏差有一定容忍度
inst_level_strict：按每条指令计算准确率（严格）
inst_level_loose：按每条指令计算准确率（宽松）
主要指标为 prompt_level_strict
支持自动验证指令遵循情况

属性

属性	值
基准测试名称	`ifeval`
数据集ID	opencompass/ifeval
论文	N/A
标签	`InstructionFollowing`
指标	`prompt_level_strict`, `inst_level_strict`, `prompt_level_loose`, `inst_level_loose`
默认示例数（Shots）	0-shot
评估划分	`train`

数据统计

指标	值
总样本数	541
提示词长度（平均）	210.75 字符
提示词长度（最小/最大）	53 / 1858 字符

样例示例

子集：default

{
  "input": [
    {
      "id": "cb71907f",
      "content": "Write a 300+ word summary of the wikipedia page \"https://en.wikipedia.org/wiki/Raymond_III,_Count_of_Tripoli\". Do not use any commas and highlight at least 3 sections that has titles in markdown format, for example *highlighted section part 1*, *highlighted section part 2*, *highlighted section part 3*."
    }
  ],
  "target": "",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "key": 1000,
    "prompt": "Write a 300+ word summary of the wikipedia page \"https://en.wikipedia.org/wiki/Raymond_III,_Count_of_Tripoli\". Do not use any commas and highlight at least 3 sections that has titles in markdown format, for example *highlighted section part 1*, *highlighted section part 2*, *highlighted section part 3*.",
    "instruction_id_list": [
      "punctuation:no_comma",
      "detectable_format:number_highlighted_sections",
      "length_constraints:number_words"
    ],
    "kwargs": [
      {
        "num_highlights": null,
        "relation": null,
        "num_words": null,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      },
      {
        "num_highlights": 3,
        "relation": null,
        "num_words": null,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      },
      {
        "num_highlights": null,
        "relation": "at least",
        "num_words": 300,
        "num_placeholders": null,
        "prompt_to_repeat": null,
        "num_bullets": null,
        "section_spliter": null,
        "num_sections": null,
        "capital_relation": null,
        "capital_frequency": null,
        "keywords": null,
        "num_paragraphs": null,
        "language": null,
        "let_relation": null,
        "letter": null,
        "let_frequency": null,
        "end_phrase": null,
        "forbidden_words": null,
        "keyword": null,
        "frequency": null,
        "num_sentences": null,
        "postscript_marker": null,
        "first_word": null,
        "nth_paragraph": null
      }
    ]
  }
}

提示模板

未定义提示模板。

Category	Subset	Metric	Score	Num
default	default	mean_prompt_level_strict	0.8000	10

IFEval · Subset Scores

IQuiz

0.8000

Description

Overview

IQuiz is a Chinese benchmark for evaluating AI models on intelligence quotient (IQ) and emotional quotient (EQ) questions. It tests logical reasoning, pattern recognition, and social-emotional understanding through multiple-choice questions.

Task Description

Task Type: Multiple-Choice Question Answering
Input: Question in Chinese with multiple choice options
Output: Selected answer with explanation (Chain-of-Thought)
Language: Chinese

Key Features

Dual evaluation of IQ and EQ capabilities
Chinese-language cognitive assessment
Multiple difficulty levels
Requires explanation alongside answer selection
Tests logical reasoning and emotional understanding

Evaluation Notes

Default configuration uses 0-shot evaluation
Primary metric: Accuracy
Subsets: IQ (logical reasoning) and EQ (emotional intelligence)
Uses Chinese Chain-of-Thought prompt template
Evaluates on test split
Metadata includes difficulty level information

Properties

Property	Value
Benchmark Name	`iquiz`
Dataset ID	AI-ModelScope/IQuiz
Paper	N/A
Tags	`Chinese`, `Knowledge`, `MCQ`
Metrics	`acc`
Default Shots	0-shot
Evaluation Split	`test`

Data Statistics

Metric	Value
Total Samples	120
Prompt Length (Mean)	248.31 chars
Prompt Length (Min/Max)	146 / 394 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
`IQ`	40	194.5	146	323
`EQ`	80	275.21	219	394

Sample Example

Subset: IQ

{
  "input": [
    {
      "id": "748f2700",
      "content": "回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式：\"答案：[LETTER]\"（不带引号），其中 [LETTER] 是 A,B,C,D 中的一个。请在回答前进行一步步思考。\n\n问题：天气预报说本周星期三会下雨，昨天果然下雨了，今天星期几？\n选项：\nA) 星期一\nB) 星期二\nC) 星期三\nD) 星期四\n"
    }
  ],
  "choices": [
    "星期一",
    "星期二",
    "星期三",
    "星期四"
  ],
  "target": "D",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "level": 1
  }
}

Prompt Template

Prompt Template:

回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 {letters} 中的一个。请在回答前进行一步步思考。

问题：{question}
选项：
{choices}

概述

IQuiz 是一个中文基准测试，用于评估 AI 模型在智商（IQ）和情商（EQ）问题上的表现。它通过多项选择题来测试逻辑推理、模式识别以及社会情感理解能力。

任务描述

任务类型：多项选择题问答
输入：中文问题及多个选项
输出：所选答案及解释（思维链，Chain-of-Thought）
语言：中文

主要特点

同时评估 IQ 和 EQ 能力
中文认知能力测评
多种难度级别
要求在选择答案的同时提供解释
测试逻辑推理与情感理解能力

评估说明

默认配置使用 0-shot 评估
主要指标：准确率（Accuracy）
子集：IQ（逻辑推理）和 EQ（情感智能）
使用中文思维链（Chain-of-Thought）提示模板
在 test 数据划分上进行评估
元数据包含难度级别信息

属性

属性	值
基准测试名称	`iquiz`
数据集 ID	AI-ModelScope/IQuiz
论文	N/A
标签	`Chinese`, `Knowledge`, `MCQ`
指标	`acc`
默认示例数	0-shot
评估数据划分	`test`

数据统计

指标	值
总样本数	120
提示词长度（平均）	248.31 字符
提示词长度（最小/最大）	146 / 394 字符

各子集统计数据：

子集	样本数	提示平均长度	提示最小长度	提示最大长度
`IQ`	40	194.5	146	323
`EQ`	80	275.21	219	394

样例示例

子集: IQ

{
  "input": [
    {
      "id": "748f2700",
      "content": "回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式：\"答案：[LETTER]\"（不带引号），其中 [LETTER] 是 A,B,C,D 中的一个。请在回答前进行一步步思考。\n\n问题：天气预报说本周星期三会下雨，昨天果然下雨了，今天星期几？\n选项：\nA) 星期一\nB) 星期二\nC) 星期三\nD) 星期四\n"
    }
  ],
  "choices": [
    "星期一",
    "星期二",
    "星期三",
    "星期四"
  ],
  "target": "D",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "level": 1
  }
}

提示模板

提示模板：

回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：[LETTER]"（不带引号），其中 [LETTER] 是 {letters} 中的一个。请在回答前进行一步步思考。

问题：{question}
选项：
{choices}

Category	Subset	Metric	Score	Num
default	IQ	mean_acc	1.0000	10
default	EQ	mean_acc	0.6000	10

IQuiz · Subset Scores

Overview

Results by Dataset

Overview

Task Description

Key Features

Evaluation Notes

Properties

Data Statistics

Sample Example

Prompt Template

概述

任务描述

主要特点

评估说明

属性

数据统计

样例示例

提示模板

Overview

Task Description

Key Features

Evaluation Notes

Properties

Data Statistics

Sample Example

Prompt Template

概述

任务描述

主要特点

评估说明

属性

数据统计

样例示例

提示模板

Overview

Task Description

Key Features

Evaluation Notes

Properties

Data Statistics

Sample Example

Prompt Template

概述

任务描述

主要特点

评估说明

属性

数据统计

样例示例

提示模板

Overview

Task Description

Key Features

Evaluation Notes

Properties

Data Statistics

Sample Example

Prompt Template

概述

任务描述

主要特点

评估说明

属性

数据统计

样例示例

提示模板