FineWeb2 Edu Japanese delivers ~120 million high‑quality educational Japanese texts (≈89.3 billion tokens) from FineWeb2, filtered by a DeepSeek‑API classifier (score ≥ 2.5), tokenized via ModernBERT‑Ja‑130M, and includes a small‑token subset (≤512 tokens).
Description in other languages
Statistics
Statistics as of Nov 16, 2026
Hugging Face downloads (last 30 days)
8,000
Hugging Face likes
36
Listed
Feb 25, 2025
Downloads (30-day window) over time
Show as a table
| Month | DL (30d) |
|---|---|
| Dec 2025 | 6,000 |
| Jan 2026 | 1,000 |
| Feb 2026 | 2,000 |
| Mar 2026 | 4,000 |
| Apr 2026 | 2,000 |
| May 2026 | 3,000 |
| Jun 2026 | 12,000 |
| Jul 2026 | 9,000 |
| Aug 2026 | 7,000 |
| Sep 2026 | 3,000 |
| Oct 2026 | 3,000 |
| Nov 2026 | 8,000 |
Downloads over the 30 days before the last snapshot of each month.
Popular in the same category
- KakologArchivesKakologArchives74kAggregates over 150 GB of NicoNico Live’s comment logs from 2009‑2024—including pre‑transition, post‑transition, and real‑time NX‑Jikkyo captures—providing an API for easy retrieval of historical TV‑broadcast discussions.
- emilia-yodasTTS-AGI18kDataset of dialogues and lore from the Fate/Stay Night character “Emilia”, formatted for training and evaluating conversational language models.
- ClimbLab-JaKantaHayashiAI15kClimbLab‑Ja is a 300 billion‑token Japanese corpus, filtered from LLM‑jp Corpus v4 into 20 clusters for pre‑training language models, released under ODC‑BY with original source licenses and stored in parquet format.
- llm-course-final-2026bumbuboon13kJapanese Wikipedia‑based RAG dataset for the LLM course final competition, containing ~1 million text chunks and a search index.
- fgo_voices_jpdeepghs10kJP Voice‑Text Dataset for FGO Waifus: a 30,800‑record, 66.4‑hour Japanese audio collection of single‑voice‑actor character lines (≈7.76 s each) useful for ASR/ASV fine‑tuning and evaluation.
- Cauldron-JAturing-motors7kCauldron‑JA is a Japanese vision‑language dataset of 44 sub‑datasets translated from The Cauldron using the DeepL API, available via HuggingFace’s datasets library and licensed identically to the original set, with prompts released under CC‑BY‑4.0.
Resources in similar research fields
- wrimeshunk031323The WRIME dataset is a Japanese collection of 42,200 posts annotated with Plutchik’s eight emotions for the writer, three readers, and their averages, structured into 40 k‑train, 1.2 k‑validation, and 2 k‑test splits for sentiment‑analysis tasks.Shared: Information Extraction & Text Mining・Annotation and Dataset Development
- cc100-jarange3265cc100-ja is a collection of the Japanese portion of the cc100 dataset, provided as sharded Parquet files.Shared: Information Extraction & Text Mining・Annotation and Dataset Development
- financial-lakehouseYoshi-Dai5kA gated, non‑commercial derivative dataset of EDINET XBRL financial data that prohibits redistribution, AI training, and commercial use, requiring manual approval for access.Shared: Annotation and Dataset Development
- Nemotron-Personas-Japannvidia5kNemotron‑Personas‑Japan is an open‑source, CC BY 4.0 dataset of high‑quality, synthetically generated Japanese personas—incorporating name, gender, age, background, marital status, education, occupation and location—grounded in real‑world demographic, geographic and personality distributions, engineered with probabilistic graphical models and GPT‑OSS‑120B to enhance diversity, reduce bias, prevent model collapse, assist sovereign AI development, and support commercial use.Shared: Annotation and Dataset Development
- mc4-jaizumi-lab2kDataset card for the Japanese MC4 dataset (mc4-ja).Shared: Annotation and Dataset Development
- RyokoAI_Syosetu711Kbotp2kSyosetu711K is a Japanese dataset of ~711,700 novels scraped from 小説家になろう on March 26‑27 2023, providing full text and metadata (title, author, NCode, synopsis, etc.) for unsupervised text generation and classification tasks.Shared: Annotation and Dataset Development