For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

kv_cache_store_ragged

def kv_cache_store_ragged[cache_t: KVCacheT, //, target: StringSlice[ImmStaticOrigin], input_fn: def[width: Int, alignment: Int](idx: IndexList[Int(3)]) capturing thin -> SIMD[cache_t.dtype, width]](cache: cache_t, input_shape: IndexList[Int(3)], input_row_offsets: LayoutTensor[DType.uint32, element_layout=input_row_offsets.element_layout, layout_int_type=input_row_offsets.layout_int_type, linear_idx_type=input_row_offsets.linear_idx_type, masked=input_row_offsets.masked, alignment=input_row_offsets.alignment], context: DeviceContext)

Stores ragged input values into a paged KV cache via an elementwise kernel.

Invokes the supplied input_fn to load values and writes them into the cache at positions determined by the per-batch cache lengths and input row offsets.

Parameters:

cache_t (KVCacheT): The KV cache type used to store key or value entries (inferred).
target (StringSlice[ImmStaticOrigin]): Compilation target string used to dispatch GPU versus CPU paths.
input_fn (def[width: Int, alignment: Int](idx: IndexList[Int(3)]) capturing thin -> SIMD[cache_t.dtype, width]): Compile-time callback that loads a SIMD vector of cache-typed elements at the given 3D index.

Args:

cache (cache_t): The KVCache object to write key or value entries into.
input_shape (IndexList[Int(3)]): Shape of the input as a 3D index list (tokens, heads, head_size).
input_row_offsets (LayoutTensor[DType.uint32, element_layout=input_row_offsets.element_layout, layout_int_type=input_row_offsets.layout_int_type, linear_idx_type=input_row_offsets.linear_idx_type, masked=input_row_offsets.masked, alignment=input_row_offsets.alignment]): Tensor with shape (batch_size + 1,) denoting the start of each sequence along the ragged sequence dimension.
context (DeviceContext): The call context pointer, passed by the graph compiler.