YCWTG/Better-Ruozhiba-ChatML

---
language:
- zh
tags:
- chatml
- sft
- chinese
- dialogue
task_categories:
- text-generation
size_categories:
- 1K<n<10K
pretty_name: Better-Ruozhiba-ChatML
license: other
---

# Better-Ruozhiba-ChatML

This dataset is a ChatML-style conversion of `FunnySaltyFish/Better-Ruozhiba`, prepared for supervised fine-tuning (SFT).

## Data Source

- Upstream dataset: `FunnySaltyFish/Better-Ruozhiba`
- This repo keeps the same content and rewrites each sample into a `messages` conversation format.

## Format

Main file: `train.json`

Each record has the following schema:

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

## Statistics

- Samples: 1,475
- Language: Chinese (`zh`)
- Average user message length: ~25 characters
- Average assistant message length: ~84 characters

## Intended Use (SFT)

- Single-turn chat SFT
- Instruction-following / conversational tuning with ChatML-compatible pipelines

## Loading Example

```python
from datasets import load_dataset

ds = load_dataset("YCWTG/Better-Ruozhiba-ChatML", split="train")
print(ds[0]["messages"])
```

## Notes

- This dataset is for research and model training purposes.
- Please check and follow the original upstream repository's terms, attribution requirements, and any license constraints before commercial use.