MiMo-V2.5
Xiaomi--attention-backend fa3 --mm-attention-backend fa3
usedinference servingin MiMo-V2.5Xiaomi
1 more from this model
--attention-backend fa3
optionalinference servingin MiMo-V2.5Xiaomi
specific method · filed under inference & serving
The third-generation FlashAttention implementation selected as an attention backend.
The sources on this page name it without explaining how it works. The line above is the extraction's one-line gloss. See inference kernel for the methods filed beside it.
Also called --attention-backend fa3, FlashAttention 3 (fa3).
2 quotes from 2 sources, one card per model, each opening with its strongest.
--attention-backend fa3 --mm-attention-backend fa3
--attention-backend fa3