---
language:
- ky
library_name: transformers
pipeline_tag: token-classification
base_model: xlm-roberta-base
tags:
- kyrgyz
- ky
- baseline
- xlm-r-base
- kyrgyz-ner
license: other
---

# Kyrgyz NER XLM-R Base

This repository contains a fine-tuned XLM-R Base checkpoint for named entity recognition on native Kyrgyz news text.

## Dataset

- Dataset: [Kyrgyz NER](https://huggingface.co/datasets/metinovadilet/kyrgyz-ner)
- Language: Kyrgyz (`ky`)

## Model

- Base model: `xlm-roberta-base`
- Role: multilingual baseline/reference checkpoint
- Framework: Hugging Face Transformers

## Reported Results

| Metric | Score |
|---|---:|
| Entity F1 | 0.5585 |
| Precision | 0.5523 |
| Recall | 0.5649 |

## Training Summary

| Setting | Value |
|---|---:|
| Epochs | 5 |
| Batch size | 128 |
| Learning rate | 2e-5 |
| Training time | 68.0 seconds |

## Notes

This is a multilingual reference checkpoint used for comparison with KyrgyzBERT and KyrgyzBERTv2. Stronger published XLM-R-style KyrgyzNER systems exist; this checkpoint is a local reference run.

## Intended Use

This checkpoint is intended for baseline/reference evaluation for Kyrgyz named entity recognition. It is intended for research, reproducibility, and educational use by the Kyrgyz NLP community. It should not be used for high-stakes decisions or production deployment without separate validation for the target domain.

## License and Usage

License metadata is set to `other`. The checkpoint is released for research and reproducibility. Downstream datasets and base models may have their own licenses or usage terms; users are responsible for following the corresponding dataset cards and upstream model licenses. The checkpoint is provided without warranty.