Skip to content

release: tha-g2p-base-1.0 (gated ByT5-base student, parity 0.0pp) - #16

Merged
ronaldtse merged 2 commits into
mainfrom
release/tha-g2p-base-1.0
Aug 20, 2026
Merged

release: tha-g2p-base-1.0 (gated ByT5-base student, parity 0.0pp)#16
ronaldtse merged 2 commits into
mainfrom
release/tha-g2p-base-1.0

Conversation

@ronaldtse

Copy link
Copy Markdown
Contributor

Summary

  • First distilled model release: tha-g2p-base-1.0 — ByT5-base student from the B-K/umt5 teacher (4.43% PER), sequence-level KD over 48,757 beam-4 labels
  • Gate: student 9.19% PER, +4.76pp ≤ +5pp budget (1,219 Kaikki sentences, beam-4, corpus-level PER)
  • Parity: ONNX KV vs torch, 1,219 samples, delta 0.0pp, zero mismatches
  • GH Release tha-g2p-base-1.0 carries the split parts (2.78GB > 2GiB cap); this PR adds the models.yaml entry + metadata

Test plan

  • CI green
  • production fetch verify (chained after merge)

@ronaldtse
ronaldtse force-pushed the release/tha-g2p-base-1.0 branch from 4dc55c9 to 91fd672 Compare August 19, 2026 20:21
@ronaldtse
ronaldtse force-pushed the release/tha-g2p-base-1.0 branch from 9143012 to f77d23d Compare August 20, 2026 12:21
@ronaldtse
ronaldtse merged commit 715917f into main Aug 20, 2026
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant