Mojo struct
KVCacheScalesMHAOperand
@register_passable(trivial)
struct KVCacheScalesMHAOperand[cache_t: KVCacheT]
An MHAOperand that accesses the scales field of a KVCache.
This is useful for MLA attention where k_s (per-token scales) are stored in the scales field of the k cache with quantization_granularity = head_size. The scales have shape [num_blocks, page_size, num_heads, head_dim_granularity].
Fields
- cache (
cache_t):
Implemented traits
AnyType,
Copyable,
DevicePassable,
ImplicitlyCopyable,
ImplicitlyDestructible,
MHAOperand,
Movable,
RegisterPassable,
TrivialRegisterPassable
comptime members
__copy_ctor_is_trivial
comptime __copy_ctor_is_trivial = cache_t.__copy_ctor_is_trivial
__del__is_trivial
comptime __del__is_trivial = cache_t.__del__is_trivial
__move_ctor_is_trivial
comptime __move_ctor_is_trivial = cache_t.__move_ctor_is_trivial
device_type
comptime device_type = KVCacheScalesMHAOperand[cache_t]
dtype
comptime dtype = cache_t.scale_dtype
page_size
comptime page_size = cache_t.page_size_
quantization_enabled
comptime quantization_enabled = cache_t.quantization_enabled
quantization_granularity
comptime quantization_granularity = cache_t.quantization_granularity
scale_dtype
comptime scale_dtype = DType.invalid
Methods
__init__
__init__(cache: cache_t) -> Self
get_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) -> UnsafePointer[Scalar[KVCacheScalesMHAOperand[cache_t].dtype], MutAnyOrigin]
Returns:
scales_block_paged_ptr
scales_block_paged_ptr(self, batch_idx: Int, start_tok_idx: Int, head_idx: Int, head_dim_idx: Int = 0) -> UnsafePointer[Scalar[DType.invalid], MutAnyOrigin]
Returns:
load_scale
load_scale[width: Int](self, batch_idx: Int, start_tok_idx: Int, head_idx: Int, head_dim_idx: Int) -> SIMD[DType.invalid, width]
Returns:
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[swizzle_mode: TensorMapSwizzle, *, BN: Int, depth: Int, BK: Int = padded_depth[KVCacheScalesMHAOperand[cache_t].dtype, swizzle_mode, depth]()](self, ctx: DeviceContext, out tma: TMATensorTile[KVCacheScalesMHAOperand[cache_t].dtype, 3, _padded_shape[3, KVCacheScalesMHAOperand[cache_t].dtype, IndexList(BN, 1, BK, Tuple()), swizzle_mode](), _ragged_shape[3, KVCacheScalesMHAOperand[cache_t].dtype, IndexList(BN, 1, BK, Tuple()), swizzle_mode]()])
TMA not supported for KVCacheScalesMHAOperand.
Returns:
create_ragged_tma_tile
create_ragged_tma_tile[swizzle_mode: TensorMapSwizzle, *, BN: Int, depth: Int, BK: Int = padded_depth[KVCacheScalesMHAOperand[cache_t].dtype, swizzle_mode, depth]()](self, ctx: DeviceContext, out tma: RaggedTMA3DTile[KVCacheScalesMHAOperand[cache_t].dtype, swizzle_mode, BN, BK])
TMA not supported for KVCacheScalesMHAOperand.
Returns:
scales_raw_ptr
scales_raw_ptr(self) -> UnsafePointer[Float32, MutAnyOrigin]
Returns a null pointer. KVCacheScalesMHAOperand already points to the scales pointer.
Returns:
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!