# PolyGraph captures one CUDA graph for adaptive speculative decoding by writing a 28-byte shape plan before each replay

On a single H800, the prototype replaced 64 shape-specific graphs, matched warmed-baseline latency, and cut graph-build time before serving from 3,915 ms to 93 ms, though only on synthetic traces.

- Published: 2026-10-20T06:12:48.347Z
- Canonical: https://polylog.news/ai/2026-10-20/polygraph-captures-one-cuda-graph-for-adaptive-speculative-d
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.LG](https://arxiv.org/abs/2610.23861), [Polylog editors](https://polylog.news)

Adaptive speculative decoding changes the active batch size B and the draft length d from one decode step to the next. A CUDA graph replays a fixed launch configuration. That mismatch forces serving systems into one of three costly choices…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-20/polygraph-captures-one-cuda-graph-for-adaptive-speculative-d (subscription information: https://polylog.news/pricing).