Xiaomi-Robotics-1
Xiaomiattn_implementation="flash_attention_2",
usedunclearin Xiaomi-Robotics-1-RoboCasa365Xiaomi Robotics
specific method · filed under inference & serving
The second-generation FlashAttention implementation selected as an attention backend.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See inference kernel for the methods filed beside it.
Also called FlashAttention 2 attention implementation, attn_implementation="flash_attention_2".
2 quotes from 2 sources, one card per model, each opening with its strongest.
attn_implementation="flash_attention_2",
"attn_implementation="flash_attention_2", # [Optional] Requires `flash_attn`