Repository navigation
Conversation
Retrain the default Thai model (`budoux/models/th.json`) on the Wisesight word-tokenization corpus using the Active-Hessian & Zero-Drift LogitBoost pipeline (`it=6000`). Also update `scripts/prepare_wisesight.py` and `scripts/run_training_pipeline.py` to: - Strip Unicode Cf invisible format characters (such as U+200B), extended pictographs, emoji variation selectors, skin-tone modifiers, and attached combining marks. - Filter out unsegmented sentences without separator boundaries and collapse consecutive whitespace chunks. - Support deterministic 80:10:10 train/val/test dataset splitting (`--split-dir`) and automatic Wisesight corpus fetching when `--lang th` is passed. Change-Id: I31efb5466e79291880d561deef944ac8c6c4708c
Change-Id: I85e65f415ce26c5379390e91f068ddfc4177b781
tushuhei
force-pushed
the
feat/retrain-th-wisesight
branch
from
October 28, 2026 02:54
a78112d to
327518c
Compare
kojiishi
approved these changes
Oct 28, 2026
| Returns: | ||
| Cleaned BudouX-segmented line using `utils.SEP` (`▁`). | ||
| """ | ||
| line = ''.join(ch for ch in line.strip() if unicodedata.category(ch) != 'Cf') |
Collaborator
There was a problem hiding this comment.
nit: Since you're already using regex, I think you can do this by:
line = regex.compile(r'\p{Cf}').sub('', line)This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Retrain the default Thai model (
budoux/models/th.json) on the Wisesight word-tokenization corpus using the Active-Hessian & Zero-Drift LogitBoost pipeline (it=6000).Pipeline & Dataset Changes
scripts/prepare_wisesight.py:preprocess_line()to strip UnicodeCfinvisible format characters (such asU+200B), extended pictographs, emoji variation selectors (U+FE0E/U+FE0F), skin-tone modifiers, and attached combining marks (\p{M}*).utils.SEP not in cleaned) and collapse consecutive whitespace chunks (| | |->| |).train/val/testdataset splitting (--split-dir,--val-ratio,--test-ratio,--seed=42), resolving the 99.4% train/val data leakage betweenwisesight-1000andwisesight-160.scripts/run_training_pipeline.py:--lang thto"wisesight"split prefix and automatically downloadwisesight-1000-samples-tokenised.labeland invokescripts/prepare_wisesight.py --split-dir(while keepingth_wordrouted to Unicode corpora).scripts/tests/test_prepare_wisesight.pyandscripts/tests/test_run_training_pipeline.py, and updatescripts/README.md.Evaluation Summary (Wisesight 80:10:10 Split & External Unicode
thBenchmark)th(3,308) F1 / Chunk F1it=10000, 80% split)it=100000, 80% split)it=6000, 80% split)mainth.json(trained on 100% of Wisesight)