NeuralVulture/ai-concepts-qa

---
license: apache-2.0
language:
- en
tags:
- conversational
- sft
- instruction-following
- machine-learning
- llama
pretty_name: Conversational QA for technical tutoring (SFT)
size_categories:
- n<1K
---

# Conversational QA — supervised fine-tuning (messages format)

**Repository:** `NeuralVulture/ai-concepts-qa`
**Dataset snapshot tag:** `v1`
**Generated:** 2026-03-25 (UTC)

## Summary

This dataset contains multi-turn chat examples for supervised fine-tuning (SFT) of instruction-tuned LLMs. Each row is a **three-message** conversation:

1. **system** — fixed tutor persona and style rules
2. **user** — a natural technical question (ML / DL / LLMs / RL)
3. **assistant** — a long-form reference answer

The JSONL was produced locally from curated Q&A pairs (`question` / `response`), then normalized for TRL / Unsloth-style `messages` training.

## Structure

- **Config:** default (`train` split)
- **Columns:** `messages` (list of `{role, content}` dicts)

## Statistics

- **Examples in this revision:** 400

## Source

- **Local export path (reference):** `/Users/rion/Desktop/exp/myself/data/training/qa_sft_messages.jsonl`

## Versioning

Releases are marked with **git tags** on this dataset repo (e.g. `v1`).
For a new data drop, re-run the upload script with a new `--version-tag` (e.g. `v2`).

## License & responsibility

Text was generated for study / tutoring use. You are responsible for compliance with the licenses and policies of any downstream models (e.g. Llama) and for how you use or redistribute this data.

## Citation

If you use this dataset, please cite the dataset repo:

```bibtex
@misc{NeuralVulture_ai_concepts_qa_v1,
title = {Conversational QA SFT Dataset (v1)},
author = {Hugging Face Hub (NeuralVulture/ai-concepts-qa)},
howpublished = {\url{https://huggingface.co/datasets/NeuralVulture/ai-concepts-qa}},
year = {2026},
}
```