naadgob/kaznlp-weights

KazNLP model weights

Weights for the KazNLP Samsung Innovation Campus capstone (Bogdan Savelyev).

Contents (models.zip, ~8.2 GB)

Path	Role
models/fasttext/fasttext_v1.bin, fasttext_v2.bin	LID baselines
models/xlm-roberta/xlm-r_v2.pt	XLM-R LID v2 (96.56% macro-F1, gold test n=461)
models/xlm-roberta/tone_v1.pt	Mixed tone v1 (2GIS)
models/tone_pretrained/ru_rubert_rureviews/	RU sentiment routing
models/tone_pretrained/kz_kazakh_sentiment_bert/	KZ sentiment routing

Install

pip install huggingface_hub
python scripts/download_kaznlp_weights.py
python scripts/setup_demo_models.py
python run_demo.py

Or download models.zip from this dataset and unzip into the project root (folder models/ must appear next to README.md).

Citation

Capstone report: docs/capstone/Final_Report.md in the GitHub repo.