Mojo struct
KVCacheMHAOperand
@register_passable(trivial)
struct KVCacheMHAOperand[cache_t: KVCacheT]
An implementation for mo.opaque KVCacheT arguments to MHA kernels.
We can eventually remove this trait and just add it as a sub-trait in the KVCacheT type, but we need to solve some cyclic dependencies first.
Fields
- cache (
cache_t):
Implemented traits
AnyType,
Copyable,
DevicePassable,
ImplicitlyCopyable,
MHAOperand,
Movable,
UnknownDestructibility
comptime members
__copyinit__is_trivial
comptime __copyinit__is_trivial = cache_t.__copyinit__is_trivial
__del__is_trivial
comptime __del__is_trivial = cache_t.__del__is_trivial
__moveinit__is_trivial
comptime __moveinit__is_trivial = cache_t.__moveinit__is_trivial
device_type
comptime device_type = KVCacheMHAOperand[cache_t]
dtype
comptime dtype = cache_t.dtype
page_size
comptime page_size = cache_t.page_size_
Methods
__init__
__init__(cache: cache_t) -> Self
get_type_name
get_device_type_name
block_paged_ptr
block_paged_ptr[tile_size: Int](self, batch_idx: UInt32, start_tok_idx: UInt32, head_idx: UInt32, head_dim_idx: UInt32 = 0) -> LegacyUnsafePointer[Scalar[KVCacheMHAOperand[cache_t].dtype]]
Returns:
LegacyUnsafePointer
cache_length
max_context_length
row_idx
row_idx(self, batch_idx: UInt32, start_tok_idx: UInt32) -> UInt32
Returns the row idx when viewing the memory as a matrix.
Returns:
create_tma_tile
create_tma_tile[BN: Int, depth: Int, swizzle_mode: TensorMapSwizzle, BK: Int = depth](self, ctx: DeviceContext, out tma: TMATensorTile[KVCacheMHAOperand[cache_t].dtype, _split_last_layout[KVCacheMHAOperand[cache_t].dtype](IndexList[3, DType.int64](BN, 1, BK, Tuple[]()), swizzle_mode, True), _ragged_desc_layout[KVCacheMHAOperand[cache_t].dtype](IndexList[3, DType.int64](BN, 1, BK, Tuple[]()), swizzle_mode)])
Creates a TMA tile for efficient GPU memory transfers.
Returns:
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!