Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding
For practitioners deploying long-context LLMs, Dustin significantly improves inference throughput without accuracy degradation, addressing a key bottleneck in speculative decoding.
Dustin addresses the verification bottleneck in speculative decoding for long-context LLMs by integrating draft model lookahead signals with target model historical attention for sparse KV cache selection, achieving 27.85x self-attention speedup and 9.17x end-to-end decoding speedup at 32k length with negligible accuracy loss.
While speculative decoding improves inference throughput for multi-batch long-context Large Language Models (LLMs), its efficiency is often limited by a verification bottleneck where Key-Value (KV) cache loading dominates latency. Existing compression methods fail in this regime: static eviction incurs accuracy loss due to saliency shift, while dynamic selection introduces prohibitive computational overhead during the verification path. We propose Dustin, a sparse verification framework designed for long-context speculative decoding. Dustin integrates lookahead signals from the draft model with historical attention from the target model to identify critical tokens with high fidelity across multi-step verification windows. To reduce recomputation latency, this approach further employs a sparse estimation scheme that restricts importance scoring to a minimal subset of attention heads. Evaluations on PG-19 and LongBench with Qwen2.5-72B demonstrate that Dustin achieves a 27.85x speedup in self-attention and a 9.17x end-to-end decoding speedup at a 32k sequence length, all with negligible accuracy degradation.