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 struct
TuningConfigSwiGLU
struct TuningConfigSwiGLU
Per-M-range tuning config for fused GEMM+SwiGLU on SM100.
N is the full output width before SwiGLU halving.
Fields
- M (
Int): - M_end (
Int): - N (
Int): - K (
Int): - mma_shape (
IndexList[Int(3)]): - cluster_shape (
IndexList[Int(3)]): - block_swizzle_size (
Int): - rasterize_order (
RasterOrder): - cta_group (
Int): - swapAB (
Bool): - k_group_size (
Int): - num_accum_pipeline_stages (
Int): - num_clc_pipeline_stages (
Int): - register_swiglu (
Bool):
Implemented traits
AnyType,
Copyable,
ImplicitlyCopyable,
ImplicitlyDeletable,
Movable,
RegisterPassable,
TrivialRegisterPassable,
TuningConfig,
Writable
Methods
__init__
def __init__(M: Int, M_end: Int, N: Int, K: Int, mma_shape: IndexList[Int(3)], cluster_shape: IndexList[Int(3)], block_swizzle_size: Int, rasterize_order: RasterOrder, cta_group: Int = Int(2), swapAB: Bool = False, k_group_size: Int = Int(1), num_accum_pipeline_stages: Int = Int(2), num_clc_pipeline_stages: Int = Int(2), register_swiglu: Bool = False) -> Self
write_to
def write_to(self, mut writer: T)
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!