Python APIs

This section covers the direct MATE Python APIs. Use these interfaces only when your framework or model architecture requires direct symbol-level integration.

Note

Check wrapper availability first. The wrapper-first workflow preserves more of the upstream package behavior and usually requires less code change.

Use direct MATE Python APIs when:

  • no wrapper matches your framework’s package surface

  • a wrapper exists but does not cover the feature you need

  • you need symbol-level control over a specific operator path

Supported API Entrypoints

Attention

Optimized entrypoints for FlashAttention, varlen, KV-cache, and MLA-related attention paths.

  • mate.flash_attn_varlen_func

  • mate.flash_attn_with_kvcache

  • mate.get_mla_metadata

  • mate.flash_mla_with_kvcache

GEMM

Low-precision GEMM entrypoints, including batched FP8 and groupwise GEMM paths.

  • mate.gemm.bmm_fp8

  • mate.gemm.gemm_fp8_nt_groupwise

Hyperconnection

Direct Hyperconnection APIs exposed through mate.hyperconnection.

KDA

Direct KDA entrypoints for fused chunked KDA when the flash_kda wrapper is not the right integration surface.

  • mate.kda.chunk_kda