Search papers, labs, and topics across Lattice.
The paper introduces MetaSD, a speculative decoding framework that leverages multiple heterogeneous drafter models to accelerate LLM inference. MetaSD dynamically allocates computational resources to these drafters by framing drafter selection as a multi-armed bandit problem, using alignment feedback from the target LLM to guide the selection process. Experiments demonstrate that MetaSD consistently outperforms single-drafter speculative decoding methods.
Forget picking just one draft model for speculative decoding – MetaSD adaptively combines multiple specialized drafters on the fly, boosting inference speed beyond what any single drafter can achieve.
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.
