X · @teortaxesTex
· X / Twitter
I can't replicate this, and maybe this is "routing to Fable" or whatever. But in theory, DeepSeek could continue the pretraining of V4s. They have int…
I can't replicate this, and maybe this is "routing to Fable" or whatever. But in theory, DeepSeek could continue the pretraining of V4s. They have intermediate checkpoints, they don't need to stick to post-train alone; and MOPD allows them to cheaply reuse domain experts anyway.永雏塔菲: 看类脑的人说v4flash api的指纹更新了,去试了下好像真的更新了