Quark-135M
A lightweight bilingual language model with 135M parameters.
Features GQA, SwiGLU, RMSNorm and RoPE. Trained on 50B+ curated tokens.
View Model →Our Models
Quark-270M
Our most powerful small model with 270M parameters.
Currently training with planned scaling to 135B tokens.
Training NowQuark-Mod
Multi-label moderation model covering 9 safety categories.
Designed for safe and practical AI deployment.
View Model →What We Focus On
⚡
Small, efficient architectures
🌍
Bilingual training from scratch
🔓
Open-source everything
💻
Real-world deployment
Resources
| 📚 Quark-135M-Bilingual | Flagship bilingual model |
| 🛡️ Quark-Mod | Content moderation model |
| 📝 HuggingFace Community | Models and datasets |
| 💻 GitHub | Code, scripts and tools |