Search papers, labs, and topics across Lattice.
The paper introduces Verification-Aware Training (VAT), a novel framework that enhances speculative decoding in large language models by simulating verification during training. By incorporating a verification head and adaptive weighting, VAT optimizes the draft model's training to reflect the sequential verification process, leading to improved performance metrics. The results show that VAT achieves up to 11.4% longer acceptance lengths and up to 8.7% faster inference times across various benchmarks, demonstrating its effectiveness in practical applications.
Speculative decoding can be significantly accelerated by training models to anticipate verification outcomes, leading to faster and more efficient inference.
Speculative decoding accelerates large language model inference by using a draft model to generate candidate tokens, which are verified by the target model in a single forward pass. Verification proceeds sequentially and discards every position from the first rejection onward, yet existing draft training relies on token-level imitation of the target with a fixed per-position weighting that reflects neither property. We introduce Verification-Aware Training (VAT), a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into supervision. VAT consists of two components: (i) a verification head, a lightweight jointly trained binary classifier that supervises the draft model on whether each position survives sequential verification; (ii) verification-adaptive weighting, which replaces the fixed weighting schedule by keeping full weight up to each sample's first rejection point and re-anchoring the decay to start there. VAT modifies only the training objective, so it can be layered on top of existing methods without changing the draft architecture, the target model, or the inference procedure. Applied to EAGLE-3 and DFlash on Qwen3-4B, Qwen3-8B, and LLaMA-3.1-8B, VAT improves average acceptance length by up to 11.4% and wall-clock speedup by up to 8.7%, with consistent gains across math, code, and chat benchmarks. Code will be available at https://github.com/naver-ai/vat
