Inference with Reference: Lossless Acceleration of Large Language Models
This addresses the computational bottleneck in LLM inference for applications where reference overlap is common, offering a lossless acceleration method.
The paper tackles the problem of slow inference in large language models by proposing LLMA, an accelerator that uses available reference texts to copy overlapping spans, achieving over 2x speed-up with identical results to greedy decoding in scenarios like search engines and conversations.
We propose LLMA, an LLM accelerator to losslessly speed up Large Language Model (LLM) inference with references. LLMA is motivated by the observation that there are abundant identical text spans between the decoding result by an LLM and the reference that is available in many real world scenarios (e.g., retrieved documents). LLMA first selects a text span from the reference and copies its tokens to the decoder and then efficiently checks the tokens' appropriateness as the decoding result in parallel within one decoding step. The improved computational parallelism allows LLMA to achieve over 2x speed-up for LLMs with identical generation results as greedy decoding in many practical generation scenarios where significant overlap between in-context reference and outputs exists (e.g., search engines and multi-turn conversations).