README.md

---
pretty_name: Natively Extended FinQA
license: cc-by-4.0
language:
  - en
task_categories:
  - question-answering
tags:
  - finance
  - finqa
  - financial-question-answering
  - numerical-reasoning
  - long-context
  - rag
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/long_train.json
      - split: validation
        path: data/long_dev.json
      - split: test
        path: data/long_test.json
      - split: private_test
        path: data/long_private_test.json
---

# Natively Extended FinQA

Natively Extended FinQA is a long-context derivative of FinQA for numerical reasoning over financial data.

It preserves the FinQA task while increasing the amount of financial context surrounding each question.

Average context increased from approximately:

`611 words per question`

to:

`5,629 words per question`

## Splits

| Hugging Face Split | File | Records |
|---|---|---:|
| `train` | `long_train.json` | 6,251 |
| `validation` | `long_dev.json` | 883 |
| `test` | `long_test.json` | 1,147 |
| `private_test` | `long_private_test.json` | 919 |

## Load with Hugging Face Datasets

```python
from datasets import load_dataset

dataset = load_dataset(
    "Mr-Rosen/Accuracy-Is-Not-Enough-FinQA-Dataset"
)

print(dataset)
print(dataset["train"][0]["qa"]["question"])
````

To load only one split:

```python
test = load_dataset(
    "Mr-Rosen/Accuracy-Is-Not-Enough-FinQA-Dataset",
    split="test",
)
```

## Load as Standard JSON

```python
import json

with open("data/long_test.json", "r", encoding="utf-8") as f:
    test = json.load(f)
```

## Dataset Structure

Important top-level fields include:

```text
id
pre_text
table
post_text
qa
```

Important `qa` fields include:

```text
question
program
exe_ans
gold_inds
```

Additional FinQA fields may also be present.

## Field Usage

### Model-Visible Full-Context Input

```text
qa.question
pre_text
table
post_text
```

### Supervised Training Target

```text
qa.program
```

### Evaluation-Only / Gold Information

Do not expose these fields to a model during normal validation or test inference:

```text
qa.program
qa.answer
qa.exe_ans
qa.gold_inds
```

Example:

```python
def extract_model_input(example):
    return {
        "question": example["qa"]["question"],
        "pre_text": example.get("pre_text", []),
        "table": example.get("table", []),
        "post_text": example.get("post_text", []),
    }
```

## FinQA Program Format

The associated experiments generate structured FinQA programs.

Example:

```json
["subtract(", "5829", "5735", ")", "EOF"]
```

Multi-step programs can reference previous operations:

```json
[
  "subtract(",
  "5829",
  "5735",
  ")",
  "divide(",
  "#0",
  "5735",
  ")",
  "EOF"
]
```

These programs should be evaluated with the FinQA evaluator.

## Dataset Construction

The expanded version adds additional native financial context around FinQA examples.

The goal is to increase context noise while retaining the original question, financial table data, program targets, execution answers, and supporting-evidence information.

Context blocks were varied so relevant evidence would not always occupy a predictable position.

Ordering within individual source blocks was preserved.

## Recommended Usage

```text
train        -> adapter training
validation   -> prompt/checkpoint/config selection
test         -> final evaluation
private_test -> private-test-style use
```

If reproducing the accompanying study, do not use the test split for tuning.

## Associated Research

The dataset supports a six-method comparison using:

`Qwen/Qwen2.5-7B-Instruct`

Methods:

* Baseline
* RAG
* LoRA
* QLoRA
* RAG + LoRA
* RAG + QLoRA

Research repository:

`MarkPaulRosenthal/Accuracy-Is-Not-Enough-Practical-Financial-QA`

GitHub dataset repository:

[MarkPaulRosenthal/Accuracy-Is-Not-Enough-FinQA-Dataset](https://github.com/MarkPaulRosenthal/Accuracy-Is-Not-Enough-FinQA-Dataset)

## Retrieval Data

The fixed practical retrieval artifacts used by the RAG-family methods are published in the companion research repository rather than this dataset repository.

They are derived model inputs, not primary benchmark splits.

## Intended Uses

Appropriate uses include:

* long-context financial QA
* numerical reasoning
* structured program generation
* retrieval evaluation
* RAG
* LoRA/QLoRA
* context-noise analysis
* full-context versus retrieved-context comparisons

## Limitations

This is a controlled FinQA-derived benchmark rather than a random sample of complete production financial documents.

Results should not automatically be generalized to arbitrary financial reports or other financial domains without external evaluation.

## License

**CC BY 4.0**

This dataset remains a derivative of FinQA and should be attributed accordingly.

## Attribution

Natively Extended FinQA builds on:

**FinQA: A Dataset of Numerical Reasoning over Financial Data**

Users should cite the original FinQA work when using this derivative dataset.