Nate@nate_512

フォロー

COLIBRI is wild. GLM-5.2, a 744-billion-parameter model, runs on a 25GB setup with zero GPU. Active params sit in RAM and the rest streams from disk on demand. It's not quick, but the responses are near-perfect on consumer hardware. Apache-2.0 and free open source, already at 24k stars on GitHub.

COLIBRI GLM-5.2 744B model

1mo

Nate@nate_512

フォロー

COLIBRI is wild. GLM-5.2, a 744-billion-parameter model, runs on a 25GB setup with zero GPU. Active params sit in RAM and the rest streams from disk on demand. It's not quick, but the responses are near-perfect on consumer hardware. Apache-2.0 and free open source, already at 24k stars on GitHub.

COLIBRI GLM-5.2 744B model

1mo