Instructions to use latincy/la_core_web_trf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- spaCy
How to use latincy/la_core_web_trf with spaCy:
!pip install https://huggingface.co/latincy/la_core_web_trf/resolve/main/la_core_web_trf-any-py3-none-any.whl # Using spacy.load(). import spacy nlp = spacy.load("la_core_web_trf") # Importing as module. import la_core_web_trf nlp = la_core_web_trf.load() - Notebooks
- Google Colab
- Kaggle
Commit History
-
Remove -any- alias: not a PEP 440 wheel filename; rejected by current pip and uv
Verified signature6e83d95 -
Release 3.9.8: wheel, -any- alias, model card, changelog
Verified signatureb6c7721
-
Update changelog for 3.9.6
Verified signature72dc4f4 -
Update model card for 3.9.6
Verified signature07f7f72 -
Add trf 3.9.6 wheel (closing-quote segmentation fix + deponent fallback)
Verified signature5ca3b44
-
fix(deps): widen latincy-preprocess pin to <0.4.0 (allow 0.3.1); dedupe spacy
Requires-Dist auto-pinned <0.3.0 at assemble time (artifact of 0.2.0 being installed); 0.3.x is API-compatible. Patched METADATA + meta.json + RECORD. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>f526517
-
v3.9.5: bundle ~110 classical abbreviations + lemmatizer punctuation fix
db29c6a
-
v3.9.4: ship updated wheel (abbreviation fix; restore .similarity())
ca26a3b
-
v3.9.4: fix abbreviation segmentation regression; restore .similarity()
The Latin tokenizer re-inherits spaCy lang/la's abbreviation exceptions, so the period after a praenomen or title (M., C., Cn., Q., Ti., Aug., cos., …) stays attached instead of detaching and ending the sentence. Consular dating formulae and name lists are no longer split into spurious sentences. Guarded by a new regression test. Also restores Token/Span/Doc .similarity() via the trf_vectors component's user hooks (cosine over transformer contextual vectors). No retraining; weights unchanged from v3.9.x. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>077886d
-
fix(v3.9.3): remove stale latincy-enclitic dep from wheel metadata
The hot patch wheel had latincy-enclitic listed as a Requires-Dist because meta.json was sourced from the installed 3.9.1 model (which predated the 188c975 vendoring commit). The enclitic_splitter factory is now vendored into functions.py — no external dep needed. Tested: virumque → uirum + que passes in clean env (no latincy-enclitic installed).cc0d6e4 -
hot patch v3.9.3: add .similarity() hook + fix enclitic splitting
- functions.py: add user_token_hooks/user_span_hooks/user_hooks["similarity"] for contextual cosine similarity via trf_vectors component - functions.py: vendor enclitic_splitter (was missing from original 3.9.3 wheel, which predated the 188c975 enclitic vendoring commit) No weights changed. No version bump.c6b1d55
-
v3.9.3: fix parenthetical segmentation; vendor token_fix/enclitic_splitter
Tokenizer re-asserts prefix/infix rules on load; token_fix and enclitic_splitter bundled (no external latincy-* deps). Weights unchanged.16972e9
-
v3.9.2: align release (token_fix + enclitic_splitter); add CHANGELOG
Config/packaging alignment across the LatinCy family; model weights unchanged.44eeba9
-
Initial release: clean wheel, correct deps (latincy-preprocess + la-latincy-lookups)
9208ad6 -
initial commit
Verified signature4c47c2f