logoalt Hacker News

kennywinkertoday at 5:27 AM1 replyview on HN

180b * 1.76 bits per weight = 39.6 gigabytes.

Best you could realistically run in 32gb is like 28gb, or a 127B param model


Replies

jokethrowawaytoday at 8:48 AM

Qwen3.8-Next, thanks to its new architecture, is quite fast even if part of it is streaming from disk