---
license: mit
task_categories:
  - question-answering
language:
  - en
tags:
  - finance
  - financial-qa
  - benchmark
  - evaluation
size_categories:
  - 1K<n<10K
configs:
  - config_name: docfinqa
    data_files:
      - split: test
        path: docfinqa/test.json
  - config_name: docmath_eval_complong
    data_files:
      - split: test
        path: docmath_eval/complong_testmini.json
  - config_name: docmath_eval_compshort
    data_files:
      - split: test
        path: docmath_eval/compshort_testmini.json
  - config_name: docmath_eval_simplong
    data_files:
      - split: test
        path: docmath_eval/simplong_testmini.json
  - config_name: docmath_eval_simpshort
    data_files:
      - split: test
        path: docmath_eval/simpshort_testmini.json
---

# FinBench Test Datasets

Financial QA test datasets for evaluating LLMs on financial reasoning.

## Quick Start

```python
from datasets import load_dataset

# Datasets with HF split support
docfinqa = load_dataset("Ayushnangia/finbench-data", "docfinqa", split="test")  # 922 items
docmath = load_dataset("Ayushnangia/finbench-data", "docmath_eval_complong", split="test")  # 300 items

# Datasets with nested structure (load as JSON)
import json
from huggingface_hub import hf_hub_download

finqa_path = hf_hub_download("Ayushnangia/finbench-data", "finqa/test.json", repo_type="dataset")
with open(finqa_path) as f:
    finqa = json.load(f)  # 1,147 items

tatqa_path = hf_hub_download("Ayushnangia/finbench-data", "tatqa/tatqa_dataset_test_gold.json", repo_type="dataset")
with open(tatqa_path) as f:
    tatqa = json.load(f)  # 277 items with answers
```

## Datasets

| Dataset | Config | Items | Method |
|---------|--------|-------|--------|
| DocFinQA | `docfinqa` | 922 | `load_dataset()` |
| DocMath (complex long) | `docmath_eval_complong` | 300 | `load_dataset()` |
| DocMath (complex short) | `docmath_eval_compshort` | 200 | `load_dataset()` |
| DocMath (simple long) | `docmath_eval_simplong` | 100 | `load_dataset()` |
| DocMath (simple short) | `docmath_eval_simpshort` | 200 | `load_dataset()` |
| FinQA | - | 1,147 | JSON download |
| TAT-QA | - | 277 | JSON download |

## CLI Download

```bash
huggingface-cli download Ayushnangia/finbench-data --local-dir data --repo-type dataset
```

## Files

```
data/
├── docfinqa/test.json                (922 items)
├── finqa/test.json                   (1,147 items)
├── tatqa/tatqa_dataset_test_gold.json (277 items with answers)
└── docmath_eval/
    ├── complong_testmini.json  (300 items)
    ├── compshort_testmini.json (200 items)
    ├── simplong_testmini.json  (100 items)
    └── simpshort_testmini.json (200 items)
```

## Sources

- [DocFinQA](https://huggingface.co/datasets/kensho/DocFinQA)
- [FinQA](https://huggingface.co/datasets/ChanceFocus/finqa)
- [TAT-QA](https://github.com/NExTplusplus/TAT-QA)
- [DocMath-Eval](https://huggingface.co/datasets/yale-nlp/DocMath-Eval)