coderian's picture
Update README.md
64accb3 verified
|
Raw History Blame Contribute Delete
1.83 kB
---
language:
- en
license: cc-by-4.0
task_categories:
- text-generation
tags:
- anime
- roleplay
- chat
- question-answering
- synthetic
pretty_name: Anime Girl Chat Dataset (English)
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Anime Girl Chat Dataset (English)
A synthetic Q&A chat dataset written in the voice of a friendly anime girl
character. Every row is a self-contained exchange: a `user` message and the
character's `anime_girl` reply.
## Dataset details
- **Rows:** 4000
- **Columns:** `user` (string), `anime_girl` (string)
- **Language:** English
- **Format:** plain text only, no emoji or kaomoji, ASCII only
- **Style layers:**
- **Sweet layer** (first half): short, warm replies of 90-230 characters using `~` and small `*action*` beats.
- **PolyBuzz-style layer** (second half): longer replies of 201-350 characters, each opening with a third-person narration block in asterisks (for example `*She looks up from the pot, steam curling around her cheeks*`) followed by warm deredere dialogue.
Note for fine-tuning: the two layers are different voice intensities. For a single
consistent persona, train on one layer or mix them deliberately.
## Splits
- `train`: 4000 rows
## Usage
```python
from datasets import load_dataset
ds = load_dataset("coderian/anime-girl-chat-dataset-en")
sample = ds["train"][0]
print(sample["user"])
print(sample["anime_girl"])
```
## Intended use
- Fine-tuning or prompt-tuning small conversational models
- Testing chat data pipelines and dataset tooling
- Character / persona experiments
## Out-of-scope use
- The character is fictional. Replies must not be presented as advice from a real person.
- Not suitable for medical, legal or safety-critical guidance.
## License
cc-by-4.0