README.md

metadata

license: mit
language:
  - ja
  - en
tags:
  - mathematics
  - synthetic
  - japanese
  - retrieval
task_categories:
  - question-answering
pretty_name: LongBook Math 2M (JA basic + advanced)
size_categories:
  - 1M<n<10M

LongBook Math 2M (Japanese basic + advanced, unified)

2,000,000 deterministic synthetic Japanese mathematics documents in one dataset: 基礎 100万件 + 高度 100万件。Built for LongBook.

• Format: JSONL, one document per line: {id, namespace, text, metadata{domain, difficulty, lang, kind, index}}
• math-*.jsonl (1,000,000): 基礎8分野 (namespace=math) — arithmetic / algebra / calculus / linear_algebra / probability / geometry / number_theory / logic_sets
• adv-*.jsonl (1,000,000): 高度8分野 (namespace=math-advanced) — analysis / ode / linear_algebra_adv / probability_adv / number_theory_adv / combinatorics_adv / vector_geometry / statistics
• text structure (Japanese): 【分野】/【問題 id】/【解法】/【答え】/【キーワード】/【確認】
• Answers are computed in Python by the generators, not hallucinated (RSA round-trip, CRT congruences, trace/det, Bayes fractions verified).
• Generators: dataset/build_math_dataset.py (--seed 42), dataset/build_advanced_math_dataset.py (--seed 123).
• Reading guide for sub-agents: dataset/MATH_SUBAGENT.md (model-agnostic, default Qwen/Qwen3.5-9B).

Usage (datasets)

from datasets import load_dataset
# all 2M
ds = load_dataset("summerMC/LongBook-math-2M", split="train", data_files="*.jsonl")
# basic only / advanced only
basic = load_dataset("summerMC/LongBook-math-2M", split="train", data_files="math-*.jsonl")
adv = load_dataset("summerMC/LongBook-math-2M", split="train", data_files="adv-*.jsonl")

Usage (LongBook retrieval)

python3 dataset/bulk_ingest_math_dataset.py --input <dir> --namespace math --db ./data/longbook.sqlite3
# namespace はファイル内の値(math / math-advanced)に合わせて --namespace を指定
curl http://127.0.0.1:30000/v1/longbook/search \
  -H 'Content-Type: application/json' \
  -d '{"query":"固有値 対角化","namespace":"math-advanced","top_k":5}'