Skip to main content

Mojo struct

ChunkedMask

struct ChunkedMask[local_window_size: Int]

Mask implementing Chunked attention.

This groups the mask into chunks of size local_window_size. Considering the following case:

  • Q_len = 7
  • K_len = 10
  • local_window_size = 4

The mask will be applied as follows: K > 0 1 2 3 4 5 6 7 8 9 Q v x--------------------x 0 | 1 1 1 1 0 0 0 0 0 0 1 | 0 0 0 0 1 1 1 1 0 0 2 | 0 0 0 0 1 1 1 1 0 0 3 | 0 0 0 0 1 1 1 1 0 0 4 | 0 0 0 0 1 1 1 1 0 0 5 | 0 0 0 0 0 0 0 0 1 1 6 | 0 0 0 0 0 0 0 0 1 1

Implemented traits​

AnyType, Copyable, DevicePassable, ImplicitlyCopyable, ImplicitlyDestructible, MHAMask, Movable, RegisterPassable, TrivialRegisterPassable

comptime members​

apply_log2e_after_mask​

comptime apply_log2e_after_mask = False

check_mask_during_decoding​

comptime check_mask_during_decoding = True

device_type​

comptime device_type = ChunkedMask[local_window_size]

mask_out_of_bound​

comptime mask_out_of_bound = True

mask_safe_out_of_bounds​

comptime mask_safe_out_of_bounds = True

Methods​

get_type_name​

static get_type_name() -> String

Returns:

String

name​

static name() -> String

Returns:

String

mask​

mask[dtype: DType, width: Int, //, *, element_type: DType = DType.uint32](self, coord: IndexList[4, element_type=element_type], score_vec: SIMD[dtype, width]) -> SIMD[dtype, width]

Returns:

SIMD[dtype, width]

status​

status[*, element_type: DType = DType.uint32](self, tile_offset: IndexList[2, element_type=element_type], tile_size: IndexList[2, element_type=element_type]) -> TileMaskStatus

Returns:

TileMaskStatus

start_column​

start_column[BM: Int, BN: Int, page_size: Int](self, row: UInt32) -> UInt32

Returns:

UInt32

total_iters​

total_iters[BM: Int, BN: Int, page_size: Int](self, row: UInt32, num_cols: UInt32) -> UInt32

Returns:

UInt32

count_nonfull_sets​

static count_nonfull_sets(BM: Int, BN: Int) -> Int

Returns:

Int

last_masked_set_end​

last_masked_set_end[BM: Int, BN: Int, page_size: Int](self, row: UInt32, num_cols: UInt32) -> UInt32

Returns:

UInt32

masked_set_ends​

masked_set_ends[BM: Int, BN: Int, page_size: Int](self, row: UInt32, num_cols: UInt32) -> StaticTuple[UInt32, ChunkedMask.count_nonfull_sets(BM, BN)]

Returns:

StaticTuple[UInt32, ChunkedMask.count_nonfull_sets(BM, BN)]

nonfull_sets​

static nonfull_sets[BM: Int, BN: Int]() -> StaticTuple[TileMaskStatus, ChunkedMask.count_nonfull_sets(BM, BN)]

Returns:

StaticTuple[TileMaskStatus, ChunkedMask.count_nonfull_sets(BM, BN)]

mask_strategies​

static mask_strategies[BM: Int, BN: Int]() -> StaticTuple[MaskStrategy, ChunkedMask.count_nonfull_sets(BM, BN)]

Returns:

StaticTuple[MaskStrategy, ChunkedMask.count_nonfull_sets(BM, BN)]