RikkaBotan's picture
Update README.md
d29cc38 verified
metadata
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:

"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.

RikkaBotan_Logo