Draft, verify, correct

An editable map of speculative decoding, with the exact-sampling correction made explicit.

Preparing the preview…

Made to be adapted.

A cheaper draft model proposes a short sequence. The target evaluates those proposals in parallel. The draft proposes; the target remains responsible for the output distribution.

Conceptual data flow, not a running language model or benchmark. Both p and q must describe the same token space and prefix.

Export formats: SVG / PNG / HTML
Editable in Studio: Explanation, appearance, and model inputs
Access: Free with attribution. No account needed.
Reference: Read the source

Loading your workspace…

Draft, verify, correct

Made to be adapted.

Another angle on the idea.

Inside a retrieval system

Draft, verify, correct

Made to be adapted.

Another angle on the idea.

Inside a retrieval system