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

rope_split_store

Fused rope + split + KV store kernel.

Reads a flat QKV matmul output, applies RoPE to Q and K regions, stores K/V to the paged KV cache, and writes roped Q to the output buffer, all in a single GPU kernel to eliminate intermediate tensor round-trips.

Functions

• ​rope_split_store_paged_ragged: Rope+split+store with paged KV cache collection.
• ​rope_split_store_paged_ragged_with_position_ids: Rope+split+store with paged KV cache and explicit position IDs.