> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tokenfactory.nebius.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How to create a dataset for fine-tuning

You can create datasets for training a model and validating the results of the training. Nebius Token Factory supports the [JSON Lines](https://jsonlines.org/) format (`.jsonl`) for dataset files. You can use one of the following dataset types:

* [Conversational](https://docs.tokenfactory.nebius.com/fine-tuning/datasets#conversational-data)
* [Instruction](https://docs.tokenfactory.nebius.com/fine-tuning/datasets#instruction-data)
* [Text](https://docs.tokenfactory.nebius.com/fine-tuning/datasets#text-data)

The size limit for dataset files is 5 gigabytes.

## **Conversational data**

You can train a model by using chats. Pass along every chat as a single `messages` line. The following is an example of how a single `messages` parameter can look:

```bash theme={null}
{
  "messages": [
    {"role": "system", "content": "This is a system prompt."},
    {"role": "user", "content": "What's for dinner tonight?"},
    {"role": "assistant", "content": "What cuisine do you prefer?"},
    {"role": "user", "content": "I wouldn't mind Italian."},
    {"role": "assistant", "content": "Then let's try a new pasta!"}
  ]
}
```

Examples of conversational datasets:

* [olathepavilion/Conversational-datasets-json](https://huggingface.co/datasets/olathepavilion/Conversational-datasets-json/blob/main/Validation.jsonl)
* [princefreddy/jsonL\_text\_to\_sql](https://huggingface.co/datasets/princefreddy/jsonL_text_to_sql/blob/main/chatbot_interactions.jsonl)

## **Instruction data**

You can specify prompts and the expected answers to them:

```bash theme={null}
{"prompt": "Capital of Australia", "completion": "Canberra"}
{"prompt": "Distance between Rome and Kuala Lumpur", "completion": "9703 km"}
```

Example of an instruction dataset: [Andzej-75/German\_RisingWorld\_prompt-text-rejected\_Jsonl](https://huggingface.co/datasets/Andzej-75/German_RisingWorld_prompt-text-rejected_Jsonl/blob/main/training_data_Lines.jsonl).

## **Text data**

If you have unstructured data, you can put each piece in a `text` line:

```bash theme={null}
{"text": "assistant: How can I help you?\n user: I need my current balance statement.\n assistant: Please find your statement in the attachment."}
{"text": "user: Can I open a bank account?\n assistant: Sure! Please fill out the following form."}
```

Example of a text dataset: [FranciscoMacaya/train\_model.jsonl](https://huggingface.co/datasets/FranciscoMacaya/train_model.jsonl/blob/main/train_model.jsonl).