wime-wanxiang

Word databases for wime, an Emacs Chinese double-pinyin input method: the rime-wanxiang pro dictionaries with Moqi auxiliary codes, keyed by double-pinyin codes, as SQLite. One file per layout:

File	Layout	Entries	sha256
wanxiang.db	Microsoft (微软双拼)	1,849,913	3dfc28ffe63b4e751b5b347097bbbc0ecd66acdc7f6c5e7fd4c274aa302ccc91
wanxiang-flypy.db	Xiaohe (小鹤双拼)	1,830,478	a6873f5f7234d7e24a2d5758a6b75eddda0ef32638dd144c3e2b135bec581d44
wanxiang-sogou.db	Sogou (搜狗双拼)	1,791,110	e14d90a1bfae8eeb3a5ae377c22368e8048f2cbd77a5bdee3190e3a16df3b480
wanxiang-ziranma.db	Ziranma (自然码)	1,830,478	8caf6e1fe1ab82a1d8a136c04e2160cb9355b39c01f5c0b1a49acc88c94a2831

Entry counts differ because each layout merges different readings onto the same keys.

Use

Download your layout's file into ~/.local/share/wime/:

uvx --from huggingface-hub hf download 0WD0/wime-wanxiang wanxiang.db \
    --repo-type dataset --local-dir ~/.local/share/wime

wanxiang.db is wime's default; for another layout set wime-database-file to that file. wime's README covers the rest of the setup.

Contents

• entry(code, word, aux, weight, tones): a word under its key code, its Moqi auxiliary groups per syllable, its frequency weight, and per syllable the tone digits (1–4, 5 neutral) of the readings behind the row.
• syllable(code, initial, final): the pinyin initial and final each pair of keys stands for; wime's neural decoder reads these, so one model serves every layout.
• meta: the layout name, the keys used in initial and final position, and syllable_encoding (wime-syllable-features-v1).

Source

rime-wanxiang dict-nightly package rime-wanxiang-moqi-fuzhu.zip, published 2026-10-30 (sha256 41122e898b5c2c3bbf2208b89e2e5ce1d4e391f0689eca50dbdb849e52ae55df), by amzxyz and contributors, CC-BY-4.0. Modified: converted with wime's gen/wime_gen.py and the layout's schemes/*.json, which map readings to double-pinyin keys and attach the auxiliary groups. Running the same command on another nightly, or with your own scheme file, builds a fresh database.

wime's evaluations used the Microsoft database; the other layouts were checked by decoding a sample sentence.