Skip to content

docs(WO88): the like-for-like khm verdict #370

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(WO88): the like-for-like khm verdict #370

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
13 changes: 13 additions & 0 deletions TODO.sota/88-secryst-migration.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,3 +27,16 @@ urd/fas lines (gates already apply to their successors).
## Acceptance
Every id in models.yaml carries recipe+corpus or an explicit
retire note; nothing remains outside the contract.


## Like-for-like verdict (2026-10-25, SAME split — secryst test, 895 rows)

- Legacy khm-latn-1.0 artifact: **CER 27.57%**
- Framework retrain (ByT5-small, 4+4 epochs, resume-tuned): **28.99%**

The migration's retrain is 1.4 points behind the legacy on identical
data — close, not beating. Next levers (recorded, in order): ByT5-base
backbone (the legacy may be base-class), corpus enrichment (merge the
Chuon Nath IPA + SBBIC vocab layers), longer schedule. The migration
contract holds: the framework path produces a honest, reproducible
28.99 where the legacy number was undocumented.
Loading