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 module

mha_decode_partition_heuristic

Split-K partition-count heuristics for MHA and MLA decode kernels.

Exposes per-backend functions that decide how many partitions the decode split-K loop should run across, plus num_keys-independent upper bounds used to launch graph-stable decode grids whose extra partitions early-return in the kernel. CUDA uses a simple SM-fill target clamped to the single-warp reducer limit; HIP (MI355X) uses a wave-aligned formula tuned for both MHA and MLA decode shapes.

Functions

• ​cuda_mha_decoding_max_num_partitions: Returns the num_keys-independent upper bound on CUDA decode split-K partitions.
• ​cuda_mha_decoding_num_partitions: Returns the CUDA decode split-K partition count for the given shape.
• ​hip_mha_decoding_num_partitions: Wave-aligned split-K target for MI355X MHA + MLA decode.
• ​mha_decoding_max_num_partitions: Returns the num_keys-independent upper bound on mha_decoding_num_partitions.
• ​mha_decoding_num_partitions: Dispatches to the backend-specific decode split-K partition heuristic.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit