Lossless Acceleration of Large Language Models with Hierarchical Drafting based on Temporal Locality in Speculative Decoding
This work addresses the problem of accelerating Large Language Model inference for real-time interactions, which is crucial for various real-world services that rely on these models.
The authors tackled the problem of accelerating inference in Large Language Models, achieving robust inference speedups with their proposed Hierarchy Drafting approach. Their method outperformed existing database drafting methods, demonstrating consistent acceleration across diverse tasks and model sizes.
Accelerating inference in Large Language Models (LLMs) is critical for real-time interactions, as they have been widely incorporated into real-world services. Speculative decoding, a fully algorithmic solution, has gained attention for improving inference speed by drafting and verifying tokens, thereby generating multiple tokens in a single forward pass. However, current drafting strategies usually require significant fine-tuning or have inconsistent performance across tasks. To address these challenges, we propose Hierarchy Drafting (HD), a novel lossless drafting approach that organizes various token sources into multiple databases in a hierarchical framework based on temporal locality. In the drafting step, HD sequentially accesses multiple databases to obtain draft tokens from the highest to the lowest locality, ensuring consistent acceleration across diverse tasks and minimizing drafting latency. Our experiments on Spec-Bench using LLMs with 7B and 13B parameters demonstrate that HD outperforms existing database drafting methods, achieving robust inference speedups across model sizes, tasks, and temperatures.