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

mla_prefill_sparse_kv_fp8

SM100 (B200) sparse MLA prefill kernel with FP8 KV cache.

Thin FP8-KV variant mirroring mla_prefill_sparse.mojo (BF16 KV). K/V latents are read as INT64-packed FP8 via SWIZZLE_NONE gather4 TMA, dequantized to BF16 in SMEM (convert_k/v_fp8_to_bf16), then fed to the shared QK/SV MMA pipeline (MLAPrefillSparseCommon.mma[fp8_active=True]). Q is BF16 in both paths, so the Q-load prologue and the MMA/softmax/correction machinery are shared through MLAPrefillSparseCommon in mla_prefill_sparse_utils.mojo.

Structs

• ​MLAPrefillSparseFP8:
• ​MLASparseSharedMemoryFP8:

Functions

• ​mla_prefill_sparse_fp8:

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