OPT-Tree: Speculative Decoding with Adaptive Draft Tree Structure

TL;DR

OPT-Tree introduces an adaptive draft tree structure for speculative decoding, significantly improving inference speed and efficiency in autoregressive language models by optimizing acceptance length during decoding.

Contribution

It proposes a novel adaptive and scalable draft tree algorithm that constructs optimal structures to maximize acceptance length, surpassing fixed heuristic methods.

Findings

Achieves up to 3.2x speed-up over autoregressive decoding.

Can generate more than ten tokens in a single step with sufficient resources.

Outperforms existing draft structures in experimental evaluations.

Abstract

Autoregressive language models demonstrate excellent performance in various scenarios. However, the inference efficiency is limited by its one-step-one-word generation mode, which has become a pressing problem recently as the models become increasingly larger. Speculative decoding employs a "draft and then verify" mechanism to allow multiple tokens to be generated in one step, realizing lossless acceleration. Existing methods mainly adopt fixed heuristic draft structures, which fail to adapt to different situations to maximize the acceptance length during verification. To alleviate this dilemma, we proposed OPT-Tree, an algorithm to construct adaptive and scalable draft trees. It searches the optimal tree structure that maximizes the mathematical expectation of the acceptance length in each decoding step. Experimental results reveal that OPT-Tree outperforms the existing draft structures and achieves a speed-up ratio of up to 3.2 compared with autoregressive decoding. If the draft model is powerful enough and the node budget is sufficient, it can generate more than ten tokens in a single step. Our code is available at https://github.com/Jikai0Wang/OPT-Tree.