v1.1.1

Documentation release for the 2026 revision: rewritten README, package description, and docs (data, protocol, reproducibility), 2026 environment record, de-identified per-name result tables, and word-token hashing in the public training and analysis scripts to match the shipped vocabulary. Model checkpoints are unchanged from v1.1.0 and are attached again for convenience; the package keeps fetching them from the v1.1.0 release.

v1.1.0

Checkpoints and results for the 2026 revision of the paper. The temporal test partition now covers names first registered from 2024 to 2026 (18,882 names, corpus 205,093). The eight seed-42 grid checkpoints attached here are the ones fetched by indonamegender.download for this version.

indonamegender v1.0.2

Release 1.0.2 of the indonamegender package on PyPI, installed with pip install indonamegender. Prediction runs on CPU and the bundled CharBiLSTM answers offline, while the other seven checkpoints download on first use from the v1.0.0 release. This version adds the baseline trainers, corrects the tokenizer path, and hardens checkpoint loading on machines without a GPU.

Package page: https://pypi.org/project/indonamegender/

Model checkpoints v1.0.0

Eight seed-42 checkpoints from the reported run, one per architecture. All eight pool through the same additive attention module. Downloaded automatically by the indonamegender package for any model other than the bundled CharBiLSTM.