Accelerating Diffusion LLMs via Adaptive Parallel Decoding

TL;DR

This paper introduces adaptive parallel decoding (APD), a method that enhances diffusion large language models' speed by dynamically adjusting token sampling, achieving higher throughput with minimal quality loss.

Contribution

The paper proposes a novel adaptive decoding method that combines diffusion LLMs with an auxiliary autoregressive model to improve decoding speed and flexibility.

Findings

APD significantly increases decoding throughput.

Minimal quality degradation observed with APD.

Flexible tradeoff between speed and quality achieved.

Abstract

The generation speed of LLMs are bottlenecked by autoregressive decoding, where tokens are predicted sequentially one by one. Alternatively, diffusion large language models (dLLMs) theoretically allow for parallel token generation, but in practice struggle to achieve the speed of autoregressive models without significantly sacrificing quality. We therefore introduce adaptive parallel decoding (APD), a novel method that dynamically adjusts the number of tokens sampled in parallel. We achieve this by defining a multiplicative mixture between the dLLM marginal probabilities and the joint probability of sequences under a small auxiliary autoregressive model. This inverts the standard setup of speculative decoding, where the goal is to sample from a large autoregressive verifier by drafting from a smaller model. We further optimize APD by enabling KV caching and limiting the size of the masked input. Altogether, our method puts forward three tunable parameters to flexibly tradeoff throughput and quality. We show that APD provides markedly higher throughput with minimal quality degradations on downstream benchmarks.