Mojo struct
ContinuousBatchingKVCacheCollection
struct ContinuousBatchingKVCacheCollection[dtype_: DType, kv_params_: KVCacheStaticParams]
This is a "view" of the cache for the given sequences in the batch.
This object does not own the underlying buffers in k_cache and v_cache, it's borrowing them from the BlockWrappers in our KVCacheManager. It does own the Pointer[NDBuffer[dtype, 3]] and valid_lengths buffer
Parameters
- dtype_ (
DType): The dtype of the kv-cache. - kv_params_ (
KVCacheStaticParams): The kv-cache static parameters.
Fields
- cache_lengths (
LayoutTensor[DType.uint32, Layout(IntTuple(-1)), ImmutAnyOrigin]): - lookup_table (
LayoutTensor[DType.uint32, Layout(IntTuple(-1)), ImmutAnyOrigin]): - blocks (
LayoutTensor[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, ContinuousBatchingKVCacheCollection[dtype_, kv_params_].blocks_layout, MutAnyOrigin]): - max_seq_length (
UInt32): - max_cache_length (
UInt32): - kv_cache_dynamic_shape (
IndexList[4]): - kv_cache_dynamic_strides (
IndexList[4]):
Implemented traits
AnyType,
Copyable,
ImplicitlyCopyable,
KVCollectionT,
Movable,
UnknownDestructibility
Aliases
__copyinit__is_trivial
comptime __copyinit__is_trivial = True
__del__is_trivial
comptime __del__is_trivial = True
__moveinit__is_trivial
comptime __moveinit__is_trivial = True
blocks_layout
comptime blocks_layout = Layout.row_major(ContinuousBatchingKVCacheCollection[dtype_, kv_params_].blocks_shape)
blocks_shape
comptime blocks_shape = IntTuple(-1, -1, -1, -1, Int(ContinuousBatchingKVCacheCollection[dtype_, kv_params_].kv_params), Int(ContinuousBatchingKVCacheCollection[dtype_, kv_params_].kv_params))
blocks_type
comptime blocks_type = LayoutTensor[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, ContinuousBatchingKVCacheCollection[dtype_, kv_params_].blocks_layout, MutAnyOrigin]
CacheType
comptime CacheType = ContinuousBatchingKVCache[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, ContinuousBatchingKVCacheCollection[dtype_, kv_params_].kv_params]
dtype
comptime dtype = dtype_
kv_params
comptime kv_params = kv_params_
name_str
comptime name_str = "continuous_batching"
Methods
__init__
__init__(out self, blocks: LayoutTensor[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, Layout.row_major[6](), MutAnyOrigin], cache_lengths: LayoutTensor[DType.uint32, Layout(IntTuple(-1)), ImmutAnyOrigin], lookup_table: LayoutTensor[DType.uint32, Layout(IntTuple(-1)), ImmutAnyOrigin], max_seq_length: UInt32, max_cache_length: UInt32)
get_key_cache
get_key_cache(self, layer_idx: Int) -> ContinuousBatchingKVCache[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, ContinuousBatchingKVCacheCollection[dtype_, kv_params_].kv_params]
Returns:
ContinuousBatchingKVCache
get_value_cache
get_value_cache(self, layer_idx: Int) -> ContinuousBatchingKVCache[ContinuousBatchingKVCacheCollection[dtype_, kv_params_].dtype, ContinuousBatchingKVCacheCollection[dtype_, kv_params_].kv_params]
Returns:
ContinuousBatchingKVCache
cache_length
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!