DocMath

Overview

DocMath-Eval is a comprehensive benchmark focused on numerical reasoning within specialized domains. It requires models to comprehend long and specialized documents and perform numerical reasoning to answer questions.

Task Description

Task Type: Document-based Mathematical Reasoning
Input: Long document context + numerical reasoning question
Output: Numerical answer with reasoning
Focus: Long-context comprehension and quantitative reasoning

Key Features

Long specialized documents requiring comprehension
Numerical reasoning within document context
Multiple complexity levels (comp/simp, long/short)
Tests real-world document understanding
Requires both reading comprehension and math skills

Evaluation Notes

Default configuration uses 0-shot evaluation
Uses LLM-as-judge for answer evaluation
Subsets: complong_testmini, compshort_testmini, simplong_testmini, simpshort_testmini
Answer format: "Therefore, the answer is (answer)"

Properties

Property	Value
Benchmark Name	`docmath`
Dataset ID	yale-nlp/DocMath-Eval
Paper	N/A
Tags	`LongContext`, `Math`, `Reasoning`
Metrics	`acc`
Default Shots	0-shot
Evaluation Split	`test`

Data Statistics

Metric	Value
Total Samples	800
Prompt Length (Mean)	68791.03 chars
Prompt Length (Min/Max)	505 / 1009038 chars

Per-Subset Statistics:

Subset	Samples	Prompt Mean	Prompt Min	Prompt Max
`complong_testmini`	300	175355.17	18687	1009038
`compshort_testmini`	200	1990.74	505	9460
`simplong_testmini`	100	13972.84	6870	24001
`simpshort_testmini`	200	3154.2	560	9600

Sample Example

Subset: complong_testmini

{
  "input": [
    {
      "id": "a07cbfcf",
      "content": "Please read the following text and answer the question below.\n\n<text>\nDELTA AIR LINES, INC.\nConsolidated Balance Sheets\n| (in millions, except share data) | March 31, 2018 | December 31, 2017 |\n| ASSETS |\n| Current Assets: |\n| Cash and cash e ... [TRUNCATED] ... comprehensive income for foreign currency exchange contracts in 2017 and 2018, and the changes in value for derivative contracts and other in 2018, in million?\n\nFormat your response as follows: \"Therefore, the answer is (insert answer here)\"."
    }
  ],
  "target": "-31.0",
  "id": 0,
  "group_id": 0,
  "metadata": {
    "question_id": "complong-testmini-0",
    "answer_type": "float"
  }
}

Note: Some content was truncated for display.

Prompt Template

Prompt Template:

Please read the following text and answer the question below.

<text>
{context}
</text>

{question}

Format your response as follows: "Therefore, the answer is (insert answer here)".

Usage

Using CLI

evalscope eval \
    --model YOUR_MODEL \
    --api-url OPENAI_API_COMPAT_URL \
    --api-key EMPTY_TOKEN \
    --datasets docmath \
    --limit 10  # Remove this line for formal evaluation

Using 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=['docmath'],
    dataset_args={
        'docmath': {
            # subset_list: ['complong_testmini', 'compshort_testmini', 'simplong_testmini']  # optional, evaluate specific subsets
        }
    },
    limit=10,  # Remove this line for formal evaluation
)

run_task(task_cfg=task_cfg)

3.7 KiB Raw Permalink Blame History