Comment

Comment by cmrdporcupine

Those two offer MoE variants, this doesn't seem to.

Dense model makes it dog slow on anything without HBM. Max 15tok/sec on decode on DDR5 systems like a Spark or a Strix Halo -- and that's at 4 bit quant.

parent cmrdporcupine 3 months ago

Discussion

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton