首页 > AI前沿 > TreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding

TreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding

arXiv自然语言 2026-08-12 13:30 6 阅读 查看原文

Speculative decoding accelerates language-model inference by letting a cheap drafter propose tokens that the target model verifies in parallel.

Recent block drafters make drafting nearly free: a single backbone pass emits an entire block of draft tokens.

Draft trees promise a further gain -- several alternative continuations verified in one target forward -- but existing constructions rank candidates by per-position marginals that ignore which parent a candidate extends, so on semi-autoregressive drafters wider trees mostly add mis-ranked nodes; and a tree of fixed size ignores how much speculation each decoding round, and each serving load, can support.

We introduce TreeSpark

TreeSpark reads a parent-conditioned distribution from the drafter's existing Markov head at negligible cost, calibrates it into an edge-acceptance estimate, and lets path survival govern everything else: best-first expansion, per-round stopping, and a load-adaptive serving policy.

Sampling siblings without replacement, with matching residuals in recursive rejection, keeps decoding lossless at any temperature.

Adaptive trees improve on matched fixed budgets at every temperature; against a tuned chain on the same drafter, TreeSpark accepts 15-25% more draft tokens per round and decodes 8-14% faster in single-request wall-clock, and under rising load it gracefully shrinks the tree back to the chain.

Code and artifacts: https://github.com/PopSoda2002/TreeSpark