[Submitted on 5 Feb 2026 (v1), last revised 23 Jul 2026 (this version, v5)]

View PDF HTML (experimental)

Abstract:Speculative decoding accelerates inference for (M)LLMs, yet a training-decoding discrepancy persists: while existing methods optimize single greedy trajectories, decoding involves verifying and ranking multiple sampled draft paths. We propose Variational Speculative Decoding (VSD), formulating draft training as variational inference over latent proposals (draft paths). VSD maximizes the marginal probability of target-model acceptance, yielding an ELBO that promotes high-quality latent proposals while minimizing divergence from the target distribution. To enhance quality and reduce variance, we incorporate a path-level utility and optimize via an Expectation-Maximization procedure. The E-step draws Monte Carlo samples from an oracle-filtered posterior, while the M-step maximizes weighted likelihood using Adaptive Rejection Weighting (ARW) and Confidence-Aware Regularization (CAR). Theoretical analysis confirms that VSD increases expected acceptance length and speedup. Extensive experiments across LLMs and MLLMs show that VSD achieves up to a 9.6% speedup over EAGLE-3 and 7.9% over ViSpec, significantly improving decoding efficiency.

Submission history

From: Xiandong Zou [view email]
[v1] Thu, 5 Feb 2026 15:36:19 UTC (79 KB)
[v2] Wed, 11 Feb 2026 09:01:01 UTC (1 KB) (withdrawn)
[v3] Thu, 12 Feb 2026 15:50:35 UTC (6 KB)
[v4] Mon, 8 Jun 2026 08:12:40 UTC (80 KB)
[v5] Thu, 23 Jul 2026 14:31:04 UTC (80 KB)