colibri
A tiny pure-C inference engine that runs frontier Mixture-of-Experts models, from 744 billion up to 2.8 trillion parameters, on consumer hardware by streaming experts from disk and treating storage, RAM and VRAM as one memory tier.
View justvugg/colibri on GitHub- GitHub stars
- 50,703
- Forks
- —
- Language
- C
Project details
- Repository
- justvugg/colibri
- Licence
- Apache-2.0
- Primary language
- C
- Catalog status
- Available
- Catalog check
- 2026-11-13
Repository description
A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.
Explore by topic
More tools in Models & training
- opencode
anomalyco · MIT
OpenCode is a free, open-source AI coding agent you run from your terminal, an alternative to tools like Claude Code or Cursor that works with any LLM provider.
- transformers
huggingface · Apache-2.0
The open-source library that provides ready-to-use implementations of virtually every major AI model for text, vision, audio, and multimodal use, letting developers run…
- comfyui
comfy-org · GPL-3.0
ComfyUI is a free, node-based visual interface for running diffusion models like Stable Diffusion locally, so you can build AI image and video generation workflows by…
- pi
earendil-works · MIT
An open-source toolkit for building AI agents: one unified API across OpenAI, Anthropic and Google models, plus a ready-made interactive coding agent CLI.