IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
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 function

ssd_chunk_map

def ssd_chunk_map[SLT: TensorLayout, SEngine: TensorEngine, QLT: TensorLayout, QEngine: TensorEngine, ROW_RULE: Bool = False](slots_i: Int32, batch_i: Int32, num_draft_tokens: Int32, info: TileTensor[.int32, SLT, MutUntrackedOrigin, Engine=SEngine], query_start_loc: TileTensor[.int32, QLT, ImmUntrackedOrigin, Engine=QEngine])

Builds the chunk table of the launch (one thread).

info holds, in order: slots chunk start tokens, slots chunk lengths (0 for an unused slot), slots owning sequences, then batch first-chunk slots and batch chunk counts. Under ROW_RULE a prefill row maps only its real tokens and a short row gets no chunk: a serial launch beside this one scans it.

Was this page helpful?