Multi-Drafter Speculative Decoding with Alignment Feedback
This addresses the problem of inefficient LLM inference acceleration for users needing diverse applications, though it appears incremental as it builds on existing speculative decoding methods.
The paper tackles the limited effectiveness of individual drafters in speculative decoding across diverse applications by introducing MetaSD, a unified framework that integrates multiple drafters and dynamically allocates computational resources using alignment feedback and multi-armed bandit selection. Experiments show MetaSD consistently outperforms single-drafter approaches.
Speculative decoding (SD) accelerates large language model (LLM) inference by using a smaller model to draft future tokens, which are then verified by the target LLM. This preserves generation quality by accepting only aligned tokens. However, individual drafters, often trained for specific tasks or domains, exhibit limited effectiveness across diverse applications. To address this, we introduce \textsc{MetaSD}, a unified framework that integrates multiple drafters into the SD process. MetaSD dynamically allocates computational resources to heterogeneous drafters by leveraging alignment feedback and framing drafter selection as a multi-armed bandit problem. Extensive experiments show MetaSD consistently outperforms single-drafter approaches.