Open Models Almanac
Methodssoftware implementationkernel & quantization library

single mention · software implementation

FLASHMLA_SPARSE

A sparse-attention backend selected by the FLASHMLA_SPARSE name.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See kernel & quantization library for the methods filed beside it.

Also called FLASHMLA_SPARSE attention backend.

Filed under software implementation › kernel & quantization library