AmareshHebbar/chinese_zodiac-sft

Chinese Zodiac (Sheng Xiao) SFT Dataset

Birth-year animal sign + element + traditional trine/clash compatibility groupings. Reuses the BaZi project's validated year-pillar logic.

Total examples	99900
Train / Validation / Test	89883 / 5052 / 4965
Question types	3
Avg citations per example	1.333
Zero-citation examples	0

Computation

Deterministic (reuses systems/bazi.py's year-pillar function directly, including the Li Chun solar-longitude year boundary — not a naive January 1st cutoff).

Validation status — read before trusting this at scale

Same validation as BaZi's year pillar: 5/5 independently-verifiable real-world year designations matched exactly (see BaZi card for the list).

Schema

Field	Type	Description
example_id	string	{synthetic_id}_{question_type}
synthetic_id	string	ID of the underlying synthetic input
system	string	"chinese_zodiac"
facts	dict	Full computed output for this system (chart/pillars/numbers depending on system)
retrieved_rules	list[dict]	Rules/observations matched for this input
question_type	string	See distribution below
user_question	string	Synthetic user question
response	string	Grounded reading — every claim traces to facts or retrieved_rules
citations	list[string]	Which rules the response draws on
source	string	"rule_matched" for all rows in this version

Question type distribution: general: 33300, compatibility_hint: 33300, yearly_traits: 33300

Usage

from datasets import load_dataset

ds = load_dataset("YOUR_USERNAME/chinese_zodiac-sft")
print(ds)

ex = ds["train"][0]
print(ex["user_question"])
print(ex["response"])

What this dataset does NOT claim

Like every dataset in this collection, this teaches a model to narrate a correctly-computed Chinese Zodiac (Sheng Xiao) reading using its own traditional rules faithfully and with citations. It does not, and cannot, contain any ground truth about whether any prediction comes true — no such data exists for any divination system. See the companion repo's top-level README for the shared "what these datasets don't claim" statement that applies across the whole collection.

Known limitations

• Only the Main animal + element — no month/day/hour-based refinements some zodiac traditions add.
• Trine/clash groupings are a simplified traditional set, not every documented compatibility rule.

Companion repo

Full generation pipeline (8 systems, shared scripts) — for regenerating at larger scale or extending the rule set — is the companion GitHub repo astrology-datasets (update this link once pushed).

License

Apache 2.0. All computed data is synthetic (no real people). Rule/effect text is written in-house, paraphrasing widely known traditional principles, not reproduced from any single copyrighted source.