IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo function

mla_mask_places_query_span

def mla_mask_places_query_span[mask_str: StaticString]() -> Bool

Returns whether the named mask derives its key spans from the query width.

Every mask but causal and null sets check_mask_during_decoding, so the decode kernel asks it for each key tile's status from a query span placed at the comptime query width. A short prompt chunk's fold pads that width up to a bucket ceiling, which would place the span past a sequence's real length, so those masks keep the decode branch's exact-width (MTP) fold only.

Parameters:

  • ​mask_str (StaticString): Mask variant name (MHAMaskVariant/MaskName spelling).

Returns:

Bool: True for every mask except "causal" and "null".

Was this page helpful?