top | item 45843378

(no title)

fspeech | 3 months ago

It uses 75% linear attention layers so it is inherently lower cost. And it is MOE so active parameters are far lower.

discuss

order

No comments yet.