Installation¶

ricelang requires Python 3.10+ and runs on Linux, macOS, and Windows. The package ships a bundled fastText detector model and 25 BPE tokenizers, totalling ~13 MB.

From PyPI¶

pipuv

pip install ricelang

uv add ricelang

This installs both the importable ricelang Python package and the ricelang command-line tool.

Verify¶

ricelang version
# 0.4.3

python -c "import ricelang as rl; print(rl.detect('မင်္ဂလာပါ'))"
# mya

What gets installed¶

Component	Size	Purpose
`ricelang/model/pdsdetect.ftz`	~1.8 MB	fastText classifier (25 trained labels)
`ricelang/model/tokenizer.crfsuite`	~1.4 MB	Burmese word-segmentation CRF
`ricelang/model/bpe_*.json` (×25)	~10 MB	per-language + multilingual BPE tokenizers
Runtime deps	varies	`fasttext-wheel`, `python-crfsuite`, `tokenizers`, `numpy<2`

Optional groups¶

ricelang's pyproject.toml declares two optional dependency groups for development / running the demo. Most users won't need them:

# to run the FastAPI demo server in demo/
uv sync --group demo

# to develop / run tests / scrape corpora
uv sync --group dev

# to build the docs site (this site)
uv sync --group docs

From source¶

git clone https://github.com/kaunghtetsan275/ricelang.git
cd ricelang
uv sync
uv run pytest

To build a wheel locally:

uv build
ls dist/

Python compatibility¶

Version	Status
3.13	⚠ untested — fasttext-wheel has no 3.13 Linux wheel yet
3.12	✓ tested in CI
3.11	✓ tested in CI
3.10	✓ tested in CI (minimum)
≤ 3.9	✗ unsupported — EOL October 2025, and patched dep versions don't backport