Automatic Task Detection and Heterogeneous LLM Speculative Decoding
This work addresses efficiency challenges in LLM inference for downstream tasks, representing an incremental improvement over existing speculative decoding methods.
The paper tackles the trade-off between acceptance rate and decoding speed in speculative decoding for LLM inference by proposing an algorithm that automatically partitions tasks and uses heterogeneous draft models aligned with task-specific data, improving draft accuracy by 6% to 50% and achieving speedups of 1.10x to 2.64x.
Speculative decoding, which combines a draft model with a target model, has emerged as an effective approach to accelerate large language model (LLM) inference. However, existing methods often face a trade-off between the acceptance rate and decoding speed in downstream tasks due to the limited capacity of the draft model, making it difficult to ensure efficiency across diverse tasks. To address this problem, we propose a speculative decoding algorithm tailored for downstream task optimization. It includes an automatic task partitioning and assigning method, which automatically categorizes downstream tasks into different sub-tasks and assigns them to a set of heterogeneous draft models. Each draft model is aligned with the target model using task-specific data, thereby enhancing the consistency of inference results. In addition, our proposed method incorporates an online lightweight prompt classifier to dynamically route prompts to the appropriate draft model. Experimental results demonstrate that the proposed method improves draft accuracy by 6% to 50% over vanilla speculative decoding, while achieving a speedup of 1.10x to 2.64x in LLM inference.