---
pretty_name: SDT GM4 Hesabu Flare Dataset v0.01
language:
- sw
license: mit
task_categories:
- text-generation
tags:
- swahili
- mathematics
- education
- instruction-tuning
size_categories:
- 1K<n<10K
---

# SDT GM4 Hesabu Flare Dataset v0.01

This dataset supports instruction fine-tuning for Swahili primary-school mathematics assistance.

## Repository contents

- `sft/train.jsonl`: training split for SFT
- `sft/val.jsonl`: validation split for SFT
- `qa/math_booster.jsonl`: targeted arithmetic booster samples

## Data format

Each row is JSON with a `messages` array in chat format, for example:

```json
{
  "messages": [
    {"role": "system", "content": "..."},
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."}
  ]
}
```

## Intended use

- Supervised fine-tuning (SFT) for Swahili math tutoring behavior
- Evaluation of response structure and arithmetic reliability

## Limitations

- Coverage is focused on school-level arithmetic and selected word problems
- Some prompts require explicit assumptions when key quantities are missing

## Safety and quality notes

- Responses should be verified for arithmetic correctness in production settings
- A deterministic post-check/guardrail is recommended for high-stakes use cases

## Citation

If you use this dataset, cite this repository:

- `Japhari/sdt-gm4-hesabu-flare-v0.01-dataset`