DIVERSED: Relaxed Speculative Decoding via Dynamic Ensemble Verification
This work addresses a specific bottleneck in accelerating large language model inference, offering an incremental improvement for researchers and practitioners in efficient AI deployment.
The paper tackles the bottleneck in speculative decoding where rigid verification reduces acceptance rates, proposing DIVERSED, a relaxed verification framework that uses an ensemble-based verifier to blend draft and target model distributions, achieving substantially higher inference efficiency.
Speculative decoding is an effective technique for accelerating large language model inference by drafting multiple tokens in parallel. In practice, its speedup is often bottlenecked by a rigid verification step that strictly enforces the accepted token distribution to exactly match the target model. This constraint leads to the rejection of many plausible tokens, lowering the acceptance rate and limiting overall time speedup. To overcome this limitation, we propose Dynamic Verification Relaxed Speculative Decoding (DIVERSED), a relaxed verification framework that improves time efficiency while preserving generation quality. DIVERSED learns an ensemble-based verifier that blends the draft and target model distributions with a task-dependent and context-dependent weight. We provide theoretical justification for our approach and demonstrate empirically that DIVERSED achieves substantially higher inference efficiency compared to standard speculative decoding methods. Code is available at: https://github.com/comeusr/diversed.