ECHO: Elastic Speculative Decoding with Sparse Gating for High-Concurrency Scenarios
This addresses the bottleneck of verification compute in production-grade LLM serving, offering a novel solution for high-concurrency scenarios.
The paper tackles the degradation of speculative decoding in high-concurrency LLM inference by introducing ECHO, a framework that reformulates it as a budgeted scheduling problem with sparse gating, achieving up to 5.35x walltime speedup and over 20% relative speedup gain.
Speculative Decoding promises to accelerate the inference of Large Language Models, yet its efficacy often degrades in production-grade serving. Existing evaluations typically overlook the compute-bound nature of high-concurrency regimes, where verification compute becomes the dominant bottleneck. Consequently, prior methods face a dilemma: static trees incur massive verification waste, while dynamic trees suffer from cumulative misjudgments and kernel incompatibility. To bridge this gap, we introduce ECHO, a high concurrency-oriented framework integrated into SGLang that reformulates speculative execution as a budgeted scheduling problem. Crucially, ECHO employs sparse confidence gating to manage the batch as a unified super-tree, elastically pivoting budget between depth and width to co-optimize the trade-off between reducing global verification steps and maximizing per-step efficiency. Extensive evaluations across diverse model scales-particularly the industrial-grade Qwen3-235B-demonstrate that ECHO consistently outperforms SOTA methods in both low-load and high-load scenarios, achieving up to 5.35x walltime speedup and delivering over 20% relative speedup gain.