For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).
Mojo module
mha_utils
comptime valuesβ
callback_fn_typeβ
comptime callback_fn_type = def[mask_t: MHAMask](mask: mask_t) raises capturing -> None
is_sm100β
comptime is_sm100 = (StringSlice("sm_100") in String(_accelerator_arch())) or (StringSlice("sm_103") in String(_accelerator_arch()))
is_sm90β
comptime is_sm90 = (StringSlice("sm_90") in String(_accelerator_arch()))
is_sm90or100β
comptime is_sm90or100 = is_sm90 or is_sm100
MHA_PDL_LEVELβ
comptime MHA_PDL_LEVEL = PDLLevel.OVERLAP_AT_END if get_defined_bool[StringSlice("MHA_PDL"), True]() else PDLLevel.OFF
Structsβ
- β
DynamicInt: - β
FlashAttentionAlgorithm: - β
MHAConfig: - β
NoPartition: - β
SplitKPartition: - β
StaticInt:
Traitsβ
Functionsβ
- β
as_dynamic_row_major_1d: - β
dispatch_mask: - β
dispatch_materialized_mask: - β
get_start_and_end_for_partitions: Calculate start and end indices for a partition.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!