README.md

---
license: apache-2.0
language:
- en
tags:
- qa
- questionanswer
- response
- chat
- llm
pretty_name: MyQA
---

# MagnumAI/MyQA

A compact conversational **prompt → response** dataset for experimenting with instruction following, supervised fine-tuning (SFT), and general-purpose conversational behavior.

The dataset contains natural-language prompts ranging from simple greetings and factual questions to everyday advice, explanations, how-to questions, and casual conversation.

## Dataset Overview

**Dataset:** `MagnumAI/MyQA`
**Format:** JSONL
**Fields:** `prompt`, `response`
**Approximate size:** 550 KB
**Task:** Conversational question answering / instruction following

Each record consists of a user-style prompt paired with an assistant response:

```json
{"prompt": "hey", "response": "Hey! 😊 What's up?"}
```

## What's Inside?

The dataset covers a range of everyday conversational patterns, including:

* Greetings and small talk
* General questions
* Geography and general knowledge
* Cooking and food
* Practical how-to questions
* Hobbies and learning
* Games and skill development
* Animal behavior
* Language and grammar
* Productivity and habits
* Entertainment recommendations
* Philosophical questions
* Casual conversation

Responses generally aim to be:

* Helpful and conversational
* Direct without being overly terse
* Structured when an explanation benefits from it
* Friendly and approachable
* Honest about uncertainty
* Actionable for practical questions

Longer responses commonly use Markdown headings, bullet points, numbered steps, and follow-up suggestions.

## Format

The dataset is stored as **JSON Lines (JSONL)**, with one example per line.

### Schema

| Field      | Type     | Description                                   |
| ---------- | -------- | --------------------------------------------- |
| `prompt`   | `string` | The user's question or conversational message |
| `response` | `string` | The corresponding assistant response          |

Example:

```json
{"prompt": "what's the capital of Australia?", "response": "Canberra! 🇦🇺 Not Sydney or Melbourne — those are the two people usually guess."}
```

Longer responses may contain escaped newlines:

```json
{
  "prompt": "how do I boil an egg?",
  "response": "Easy one — here's the reliable method.\n## The Steps\n1. Place eggs in a pot and cover with cold water..."
}
```

## Intended Uses

`MagnumAI/MyQA` can be used for:

* Supervised fine-tuning experiments
* Instruction-following experiments
* Conversational model prototyping
* Testing prompt/response formatting
* Small-scale chatbot experiments
* Studying conversational response styles
* Dataset preprocessing and JSONL pipeline testing

A typical training example can be converted into the format expected by your model or training framework:

```python
{
    "messages": [
        {"role": "user", "content": example["prompt"]},
        {"role": "assistant", "content": example["response"]}
    ]
}
```

## Limitations

This is a relatively small dataset and **should not be treated as a comprehensive knowledge or safety benchmark**.

In particular:

* The dataset is not designed to provide exhaustive factual coverage.
* Some responses contain recommendations or subjective advice.
* General-knowledge answers may become outdated over time.
* Practical advice may require additional context.
* The dataset does not provide a comprehensive evaluation of model safety.
* The examples primarily represent everyday conversational interactions rather than adversarial or safety-critical prompts.
* A model trained on this dataset may learn its particular tone, formatting habits, and response patterns.

Models fine-tuned on this dataset should be evaluated independently rather than assuming dataset quality guarantees model quality.

## Example

**Prompt**

> how do I stop procrastinating?

**Response**

> Procrastination is usually less about laziness and more about avoiding discomfort — boredom, anxiety, ambiguity, whatever the task stirs up. So the fix isn't willpower, it's lowering the friction.
>
> ## Start Smaller Than Feels Reasonable
>
> "Write the report" is too big. "Open the doc and write one sentence" is doable. The goal is to get past the starting point, because momentum often handles the rest.
>
> ...

This illustrates the dataset's general preference for **conversational explanations combined with structured, practical advice**.

## Fine-Tuning

For a chat-oriented model, examples can be represented as:

```json
{
  "messages": [
    {
      "role": "user",
      "content": "what's the capital of Canada?"
    },
    {
      "role": "assistant",
      "content": "Ottawa! 🇨🇦 Toronto and Vancouver get guessed more often, but Ottawa's the one."
    }
  ]
}
```

The exact formatting should be adapted to the chat template expected by the base model.

## Dataset Characteristics

| Characteristic   | Description                          |
| ---------------- | ------------------------------------ |
| Data type        | Conversational QA                    |
| Input            | Natural-language prompts             |
| Output           | Assistant responses                  |
| Storage          | JSONL                                |
| Size             | ~300 KB                              |
| Style            | Friendly, conversational, structured |
| Primary language | English                              |
| Intended setting | General-purpose conversational AI    |

## License

See the repository's Hugging Face dataset metadata for the applicable license.

If you are redistributing, modifying, or using the dataset commercially, verify that the license and provenance of the underlying examples permit your intended use.

## Contributions

Suggestions, corrections, additional examples, and improvements to the dataset are welcome.

When contributing examples, prefer:

* Clear prompts
* Useful, self-contained responses
* Accurate information
* Natural conversational language
* Consistent formatting
* No unnecessary personal or sensitive information

## Acknowledgements

Created and maintained by **MagnumAI**.