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).

Python function

declares_skippable_draft

declares_skippable_draft()​

max.pipelines.speculative.spec_width_policy.declares_skippable_draft(spec_config)

source

Whether a spec-decode graph takes its drafter’s row count as an input.

The drivers call this to decide whether to declare the draft_slot_ids / draft_block_offsets inputs, and the pipeline calls it to decide whether to bind them, so the two always agree and no model opts in by hand. A graph declares them only when the schedule can reach width 0, so one that never skips keeps the graph it had without them. Whether a step then runs its draft on fewer rows is the driver’s call: its draft has to declare supports_zero_draft_rows, and the driver has to skip on the graph’s topology.

Parameters:

spec_config (SpeculativeConfig | None) – The pipeline’s speculative config.

Returns:

True when some decode batch size verifies no drafts.

Return type:

bool