Positronick
← Registry
CLI Official

colibri

Pure-C inference engine that streams a 744B-parameter MoE model (GLM-5.2) from disk to run on a ~25GB-RAM consumer machine.

Install

via Positronick CLI

$positronick cli install colibri --run

Official

$git clone https://github.com/JustVugg/colibri && cd colibri/c && ./setup.sh

colibri (by JustVugg) is an experimental, dependency-free inference engine written in pure C that runs the 744B-parameter GLM-5.2 mixture-of-experts model on a consumer machine with roughly 25GB of RAM by streaming experts from disk one at a time. It includes an offline FP8-to-int4 converter and an optional CUDA tier for pinned experts, and exposes an OpenAI-compatible HTTP endpoint. Apache-2.0; build from source with the bundled setup script. It is very new and fast-moving — expect rough edges and verify model-file compatibility before relying on it.

Official source ↗
Theme