Mojo trait
KVCacheT
Trait for different KVCache types and implementations.
Represents a single (key or value) cache.
Implemented traits
AnyType,
Copyable,
ExplicitlyCopyable,
Movable,
UnknownDestructibility
Aliases
dtype
alias dtype
kv_params
alias kv_params
Methods
__copyinit__
__copyinit__(out self: _Self, existing: _Self, /)
Create a new instance of the value by copying an existing one.
Args:
- existing (
_Self): The value to copy.
Returns:
_Self
__moveinit__
__moveinit__(out self: _Self, var existing: _Self, /)
Create a new instance of the value by moving the value of another.
Args:
- existing (
_Self): The value to move.
Returns:
_Self
cache_lengths_nd
cache_lengths_nd(self: _Self) -> NDBuffer[uint32, 1, MutableAnyOrigin]
Returns the cache lengths as a NDBuffer.
Returns:
cache_length
cache_length(self: _Self, batch_idx: Int) -> Int
Returns the length of the cache for a given batch index.
Returns:
load
load[width: Int](self: _Self, bs: Int, head_idx: Int, tok_idx: Int, head_dim_idx: Int) -> SIMD[get_witness(_Self, kv_cache::types::KVCacheT, dtype), width]
Loads an element from the given index.
Returns:
store
store(self: _Self, bs: Int, head_idx: Int, tok_idx: Int, head_dim_idx: Int, val: SIMD[get_witness(_Self, kv_cache::types::KVCacheT, dtype), size])
Stores an element at the given index.
empty_cache
empty_cache(self: _Self) -> Bool
Returns true if the cache_lengths for all requests is 0, false otherwise.
Returns:
max_prompt_length
max_prompt_length(self: _Self) -> SIMD[uint32, 1]
Returns the maximum sequence length across all batches of the current request.
Returns:
max_context_length
max_context_length(self: _Self) -> SIMD[uint32, 1]
Returns the maximum cache length used across all batches of the current request.
Returns:
block_paged_ptr
block_paged_ptr[tile_size: Int](self: _Self, batch_idx: Int, start_tok_idx: Int, head_idx: Int, head_dim_idx: Int = 0) -> UnsafePointer[SIMD[get_witness(_Self, kv_cache::types::KVCacheT, dtype), 1]]
Returns a LayoutTensor pointing to the KVCache block at the given index.
Paged KVCache implementations must have a block_size which is a multiple of the and greater than the layout's first dimension.
Returns:
UnsafePointer
max_tile_size
row_idx
row_idx(self: _Self, batch_idx: SIMD[uint32, 1], start_tok_idx: SIMD[uint32, 1]) -> SIMD[uint32, 1]
Returns the row idx when viewing the memory as a matrix.
Returns:
col_idx
col_idx(self: _Self, head_idx: SIMD[uint32, 1]) -> SIMD[uint32, 1]
Returns the col idx when viewing the memory as a matrix.
Returns:
create_tma_tile
create_tma_tile[tile_m: Int, tile_n: Int, swizzle_mode: TensorMapSwizzle, *, is_k_major: Bool](self: _Self, ctx: DeviceContext) -> TMATensorTile[get_witness(_Self, kv_cache::types::KVCacheT, dtype), tile_layout_k_major[::DType,::Int,::Int,::TensorMapSwizzle]() if is_k_major else tile_layout_mn_major[::DType,::Int,::Int,::TensorMapSwizzle](), _tma_desc_tile_layout[::DType,::Int,::IndexList[$1, ::DType(), is_k_major]
Creates a TMA tile for this KV cache.
Returns:
TMATensorTile
copy
copy(self: _Self) -> _Self
Explicitly construct a copy of self.
Returns:
_Self: A copy of this value.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!