Repository navigation
tha-g2p-base-1.0
·
737 commits
to main
since this release
tha-g2p-base-1.0
IMF v1 (fp32, decoder kv, opset 14). Trained from sequence-level KD from the B-K/umt5-thai-g2p-v2-0.5k teacher (4.43% PER on this harness; secryst's saved umt5 artifacts are unusable — transformers 5.15 dropped the untied lm_head — and the volume's epitran corpus is tone-less, so the published 2.32% tier is unrecoverable until secryst regenerates it); 48,757 beam-4 teacher-generated labels; ByT5-base init google/byt5-base; checkpoint secryst-checkpoints:/secryst_thai_g2p_distill_small/run-004/best.
| field | value |
|---|---|
| task | g2p (Thai → IPA) |
| artifact | tha-g2p-base-1.0-fp32.zip (2.59 GiB) |
| parts | tha-g2p-base-1.0-fp32.zip.part-00 + tha-g2p-base-1.0-fp32.zip.part-01 (GitHub 2GiB cap) |
| per_teacher | 4.43 — beam-4, corpus-level PER (total_ed/total_gold over chars of joined-piece decode); 1,219 Kaikki Thai test sentences; B-K/umt5-thai-g2p-v2-0.5k teacher |
| per_student | 9.19 — beam-4, corpus-level PER, same harness as the teacher; gate +4.76pp <= +5pp (docs/DISTILL-SOURCE-PROMPT.md); exact match 90.81% |
| parity | cer_delta 0.0pp on 1219 samples |
| sha256 | f341a5868867a9847d3f897d1fcc4d1d63b9a59b8913871bcf4ecab48f4fa895 |
| license | BSD-3-Clause |
Runtimes reassemble split parts transparently and verify every sha256:
from interscript_ml import Model
model = Model.load("tha-g2p-base-1.0")