RikkaBotan/FineDataset_13B_JpEn

---
license: odc-by
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: text
dtype: string
splits:
- name: train
num_bytes: 55216152831
num_examples: 6103514
download_size: 28823398445
dataset_size: 55216152831
---

# Saint Iberis Pretraining Dataset
A curated multilingual corpus designed for training **Saint Iberis**, an LLM developed by **RikkaBotan** with a gentle, introspective personality and strong bilingual (JA/EN) capabilities.
This dataset blends high-quality Japanese and English web + PDF corpora with carefully chosen token ratios to balance linguistic coverage and domain diversity.

---

## 📚 Dataset Overview

The dataset is composed of four major sources:

| Key | Description | Language | Tokens |
|-----------------|-----------------------------------|----------|--------|
| fineweb2_ja | FineWeb2 (Japanese) | JA | 2.75B |
| finepdfs_ja | FinePDFs (Japanese subset) | JA | 1.00B |
| finewebedu_en | FineWeb Edu (English educational) | EN | 7.00B |
| finepdfs_en | FinePDFs (English subset) | EN | 2.25B |

**Total tokens:** **13B**

The distribution emphasizes:
- High-quality educational English web data
- Solid Japanese coverage using both web and structured PDF extractions
- Balanced domain mixture suitable for reasoning, linguistic fluency, and instruction-following

---

## 🔧 Dataset Configuration

Below are the HuggingFace dataset sources and subsets used:

```bash
"fineweb2_ja": {"hf": "hotchpotch/fineweb-2-edu-japanese", "subset": "default"}
"finepdfs_ja": {"hf": "HuggingFaceFW/finepdfs", "subset": "jpn_Jpan"}
"finewebedu_en": {"hf": "HuggingFaceFW/fineweb-edu", "subset": "sample-350BT"}
"finepdfs_en": {"hf": "HuggingFaceFW/finepdfs", "subset": "eng_Latn"}
```

# 🌸 About us

Japanese independent researcher having shy and pampered personality. Twin-tail hair is a charm point. Interested in nlp. Usually using python and C.