744B parameters. On a laptop. With 25GB RAM. Colibri runs GLM-5.2 (744B MoE) in pure C with zero dependencies. The trick: only ~40B params activate per token, so it keeps the dense part resident and streams experts from disk on demand. A single 2,400-line C file. No GPU, no BLAS, no Python at runti
@emilianobonassi·Jul 11, 2026·2 sources·positive
Read articleAI Summary
Colibri runs the 744B-parameter GLM-5.2 MoE model on a laptop with 25GB RAM using pure C and disk streaming of expert weights.
All Sources
744B parameters. On a laptop. With 25GB RAM. Colibri runs GLM-5.2 (744B MoE) in pure C with zero dependencies. The trick: only ~40B params activate per token, so it keeps the dense part resident and streams experts from disk on demand. A single 2,400-line C file. No GPU, no BLAS, no Python at runti
@emilianobonassi
Jul 11, 2026
744B parameters. On a laptop. With 25GB RAM. Colibri runs GLM-5.2 (744B MoE) in pure C with zero dependencies. The trick: only ~40B params activate per token, so it keeps the dense part resident and streams experts from disk on demand. A single 2,400-line C file. No GPU, no BLAS, no Python at runti
@jgarzik
Jul 11, 2026