logoalt Hacker News

Philpax • yesterday at 5:44 PM • 3 replies • view on HN

Are we sure this is a new pretrain and not them rounding up 2.8T?

Replies

vikramkr • yesterday at 5:53 PM

Literally from the linked announcement:

> Kimi K3.1 Pro is a post-trained version of Kimi K3, with substantially larger post-training compute

minimaxir • yesterday at 5:45 PM

The Kimi K3 announcement (https://www.kimi.ai/blog/kimi-k3) used 2.8T and it would be weird to round that up for this announcement.

➕ show 1 reply
rfoo • yesterday at 5:47 PM

it's almost certainly rounding up 2.8T