Skip to content

docs(WO66): the register arm halved the poetry gap #371

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ronaldtse merged 1 commit into from
Oct 26, 2026
Merged

docs(WO66): the register arm halved the poetry gap #371

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions TODO.sota/66-b5-persian-registers.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,3 +31,15 @@ quantified. The poetry corpus + eval are pinned
(<0.7 across registers) requires poetry in training: the 3,497
vocalized verses are the seed; the owner source picks (dossier above)
scale it.


## Register arm RESULT (2026-10-26): the poetry seed HALVED the gap

The register-adapted model (fas-diac-base fine-tuned on the full 446k
corpus + the 3,497-vocalized-verse seed x10; gate-passed 0.69% on
news-classic) scores **CER 16.55%** on the identical Ganjoor eval —
vs 33.67% pre-register. HALVED. The per-section trajectory shows the
residual concentrates in dense-vocalization verses (the late eval
sections). Scale-up path: the owner source picks (dossier) grow the
vocalized seed; the recipe (checkpoint fine-tune + seed oversample)
is proven.
Loading