744B Model Runs Locally on a Laptop
WesEklund · x · 2026-07-11
Colibri runs GLM-5.2 744B MoE on a laptop using pure C with zero dependencies, requiring only 25GB of RAM according to the author.
The core ideas:
- Each token activates only about 40B parameters, allowing the dense part to stay in memory
- Expert parameters are streamed from disk on demand, not loaded all at once
- The entire implementation is roughly 2400 lines of C code, no GPU, BLAS, or Python needed at runtime
This demonstrates a more aggressive approach to running large MoEs locally on CPU.
Related event: 744B GLM-5.2 MoE Model Runs Locally on 25GB RAM(5 posts)→
More from Infra
- Unconventional AI acquires inference chip firm NeuReality in 1,000x efficiency bid — NaveenGRao · 2026-10-27
- Musk: robot RAM demand will dwarf everything else, and 100% of superintelligence compute goes to deep space — elonmusk · 2026-10-27
- RTX 5090 hits $7K as PC gaming community goes into 'denial' over AI boom — PAstynome · 2026-10-27
- Wan2GP lands on Pinokio: 1-click AI video on 6GB GPUs — cocktailpeanut · 2026-10-27
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27