Show HN: colibri streams model weights from SSD to run GLM 5.2 on low-RAM machines

Sources: colibri repository, HN discussion
Summary: A Show HN project (colibri) runs the large GLM 5.2 model on a memory-constrained machine by streaming weights from an SSD or NVMe drive rather than holding them in RAM. The thread reached 494 points.
Comments: Commenters compare it to earlier weight-streaming work, ask about SSD write endurance, and discuss running it on GPUs with onboard NVMe slots.
Why it matters: It shows continued practitioner interest in running frontier-scale open models on commodity hardware by trading throughput for lower memory.