---
title: README
emoji: 🌟
colorFrom: yellow
colorTo: yellow
sdk: static
pinned: true
---

# About Twinkle AI

![banner-twinkle-hf](https://cdn-uploads.huggingface.co/production/uploads/618dc56cbc345ca7bf95f3cd/RW1z8a7_kxktdllwEoh3N.jpeg)

🌏 [Website](https://twinkleai.tw/) | 🎮 [Discord](https://discord.com/servers/twinkle-ai-1310544431983759450) | 😺 [GitHub](https://github.com/ai-twinkle) | 🤗 [Hugging Face](https://huggingface.co/twinkle-ai) | 🤖 [ModelScope](https://modelscope.cn/organization/twinkle-ai) | 🐳 [Docker Hub](https://hub.docker.com/u/aitwinkle) | 🔗 [Linkedin](https://www.linkedin.com/company/twinkle-ai/) | 👥 [Facebook](https://www.facebook.com/groups/twinkleai)

🌟 **Twinkle AI** is a research community founded in late 2024, dedicated to enhancing Traditional Chinese language models with local cultural context. Starting with open-source LLaMA models, we are developing practical technologies tailored to the linguistic nuances of Taiwan. Our community is made up of people from all walks of life, united by a shared passion for model training. From collecting region-specific corpora to training language models, we are committed to open-sourcing both our datasets and our models.

Our mission is to promote knowledge of large language model training through real-world practices. By embracing openness and collaboration, we aim to advance the local ecosystem in the field of generative AI.

👋 Join our [Discord](https://discord.com/servers/twinkle-ai-1310544431983759450) to connect and collaborate with the community!

### Featured Projects
- 📊 **[Twinkle Eval](https://github.com/ai-twinkle/Eval)**: A fast and accurate AI evaluation tool that uses parallel and randomized testing methods to provide objective performance analysis and stability assessments. It supports various commonly used benchmark datasets.
- 📈 **[Eval Analyzer](https://github.com/ai-twinkle/eval-analyzer)**: A comprehensive visualization and analysis tool designed to interpret evaluation logs, providing deep insights into model performance through granular error tracking and comparative reporting.