Files
aya-expanse-8b-mkd_cyrl/README.md
ModelHub XC 1e8e18cfdf 初始化项目,由ModelHub XC社区提供模型
Model: DGurgurov/aya-expanse-8b-mkd_cyrl
Source: Original Platform
2026-06-18 16:50:16 +08:00

60 lines
1.9 KiB
Markdown

---
language:
- mk
license: cc-by-nc-4.0
base_model: CohereLabs/aya-expanse-8b
tags:
- subnetwork-enhancement
- low-resource
- language-specific
pipeline_tag: text-generation
---
# Aya Expanse 8B Mkd_Cyrl
Language-enhanced Aya-Expanse-8b model for **Macedonian** using sparse subnetwork fine-tuning.
## Method
- Training approach: Language-specific neuron identification + subnetwork fine-tuning
- Parameters trained: <1% of total model parameters
- Framework: [Language Subnetwork Enhancement](https://github.com/d-gurgurov/Language-Subnetwork-Enhancement-LLMs)
## Performance
Enhanced monolingual capabilities in Macedonian while preserving multilingual performance.
## Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("DGurgurov/aya-expanse-8b-mkd_cyrl")
tokenizer = AutoTokenizer.from_pretrained("DGurgurov/aya-expanse-8b-mkd_cyrl")
prompt = "Your Macedonian prompt here"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0]))
```
## Citation
```bibtex
@misc{gurgurov2025sparsesubnetworkenhancement,
title={Sparse Subnetwork Enhancement for Underrepresented Languages in Large Language Models},
author={Daniil Gurgurov and Josef van Genabith and Simon Ostermann},
year={2025},
eprint={2510.13580},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.13580}
}
@misc{gurgurov2025languagearithmeticssystematiclanguage,
title={Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation},
author={Daniil Gurgurov and Katharina Trinley and Yusser Al Ghussin and Tanja Baeumel and Josef van Genabith and Simon Ostermann},
year={2025},
eprint={2507.22608},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2507.22608},
}
```