News for hackers. Stuff that matters.

A comment on DeepSeek-v4.1-Flash-1017. Parent comment

Re:DeepSeek-v4.1-Flash-1017 by RGS1811 on Saturday October 17, 2026 @11:34AM (#50132003)
Forgive the ignorance, but how are you running a 763B parameter model on 64GB of RAM?
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by liuliu on Saturday October 17, 2026 @12:02PM (#50132262)
SSD weights streaming. The actual weights is 552B, then engram.
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by pumanoir on Saturday October 17, 2026 @12:40PM (#50132595)
which engine?
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by liuliu on Saturday October 17, 2026 @01:41PM (#50133199)
https://releases.drawthings.ai/p/public-beta-of-local-code-b...
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by mococa on Saturday October 17, 2026 @11:38AM (#50132047)
By “quantitizing” it
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by ar-nelson on Saturday October 17, 2026 @12:00PM (#50132248)
The smallest quant is still over 100GB, so I don't know how that fits in 64GB RAM. It would have to be partly streamed from disk.
[ Parent | on HN ]