About
High-performance C++ implementation of LLM inference. Run Llama, Mistral, and other models on CPU with state-of-the-art quantization.
Features
- Extremely fast CPU inference using AVX2, NEON, and other SIMD optimizations.
- Support for 2-bit to 8-bit GGUF quantization for model size reduction.
- OpenAI-compatible API endpoint for use with custom clients like SillyTavern or Oobabooga.
- Multimodal support for models like LLaVA and BakLLaVA.
- Cross-platform support: Windows, macOS, Linux, and even iOS/Android via library integration.
Links
Categories
Reviews
4.6
Write a Review
Get new AI tools weekly
Subscribe to our newsletter and never miss a tool.
Get new AI tools weekly
Subscribe to our newsletter and never miss a tool.