Download KTransformers

Get the latest version of KTransformers for your platform

Latest Release
0.3.1
Released on 2026-01-15
All Releases

What's new:

  • Improved MoE offloading performance
  • Added support for DeepSeek-R1-671B
  • Fixed memory leak in long conversations
  • Better error messages for CUDA initialization

Choose Your Platform

🐧
Linux
Ubuntu 20.04+, CUDA 11.8+
pip install ktransformers

AppImage bundles CUDA runtime β€” no system CUDA Toolkit required, only an NVIDIA driver

Windows
Windows 10/11, CUDA 11.8+
pip install ktransformers
Quick Install
The easiest way to install KTransformers is via pip
pip install ktransformers

For detailed installation instructions including CUDA setup, see the installation guide.

System Requirements

Minimum

  • Python 3.9+
  • CUDA 11.8+ (for GPU acceleration)
  • 16 GB RAM
  • NVIDIA GPU with 8GB+ VRAM

Recommended (for large models)

  • Python 3.10+
  • CUDA 12.0+
  • 256 GB+ RAM
  • NVIDIA RTX 4090 or higher