Nate@nate_512

متابعة

Colibrì GLM-5.2 25GB RAM

The notion that large models require massive amounts of RAM is now obsolete. Colibrì, a tiny engine designed for immense models, runs GLM-5.2,a 744-billion-parameter model,on a machine with just 25GB of RAM and no GPU. By keeping only the necessary parameters in memory and streaming the rest from disk as needed, it achieves this feat efficiently.

2mo

Nate@nate_512

متابعة

Colibrì GLM-5.2 25GB RAM

The notion that large models require massive amounts of RAM is now obsolete. Colibrì, a tiny engine designed for immense models, runs GLM-5.2,a 744-billion-parameter model,on a machine with just 25GB of RAM and no GPU. By keeping only the necessary parameters in memory and streaming the rest from disk as needed, it achieves this feat efficiently.

2mo