For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo module

qsa

Graph-op bindings for the QSA lightning indexer and its attention consumer.

The kernels live in //Kernels/lib/qsa (qsa.compress_keys, qsa.block_score, qsa.sparse_attention); only the @extensibility.register wrappers are here, because a registration has to be declared inside the built-in kernel library for a served graph to resolve the op. Same shape as the attn_res.mojo and msa.mojo bindings.

Structs

• ​QSABlockScore: Relu-summed, block-causally masked QSA block scores.
• ​QSACompressKeysPaged: Pools a paged group of raw QSA index keys into dense block keys.
• ​QSASparseAttentionRaggedPaged: Gather-GQA over the token indices a QSA selection names.