For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo struct
Struct_mha_ragged_paged_sink_weights_scalar_args
struct Struct_mha_ragged_paged_sink_weights_scalar_args
Registers the attention.mha.ragged.paged.sink_weights graph op with the graph compiler.
Like attention.mha.ragged.paged, it splits a mixed step by the KV
leaf's mixed_step_split record on NVIDIA targets.
Implemented traitsโ
Methodsโ
executeโ
static def execute[mask_str: StaticString, local_window_size: Int = Int(-1), v_head_dim: Int = Int(-1)](output: T, q: T, input_row_offsets: T, kv_blocks: T, page_stride: T, cache_lengths: T, kv_lookup_table: T, max_prompt_length: T, max_cache_length: T, layer_idx: UInt32, scale: Float32, sink_weights: T, mha_decode_dispatch_metadata: T, mixed_step_split: T, context: DeviceContext)