Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time
This addresses the problem of expensive LLM inference for AI application developers, offering a novel method that is not incremental as it enables wall-clock speedups without sacrificing in-context learning ability.
The paper tackles the high computational cost of large language models (LLMs) at inference time by introducing contextual sparsity, which exploits input-dependent sparsity in attention heads and MLP parameters to speed up inference without retraining or compromising quality, achieving over 2X latency reduction compared to state-of-the-art methods.
Large language models (LLMs) with hundreds of billions of parameters have sparked a new wave of exciting AI applications. However, they are computationally expensive at inference time. Sparsity is a natural approach to reduce this cost, but existing methods either require costly retraining, have to forgo LLM's in-context learning ability, or do not yield wall-clock time speedup on modern hardware. We hypothesize that contextual sparsity, which are small, input-dependent sets of attention heads and MLP parameters that yield approximately the same output as the dense model for a given input, can address these issues. We show that contextual sparsity exists, that it can be accurately predicted, and that we can exploit it to speed up LLM inference in wall-clock time without compromising LLM's quality or in-context learning ability. Based on these insights, we propose DejaVu, a system that uses a low-cost algorithm to predict contextual sparsity on the fly given inputs to each layer, along with an asynchronous and hardware-aware implementation that speeds up LLM inference. We validate that DejaVu can reduce the inference latency of OPT-175B by over 2X compared to the state-of-the-art FasterTransformer, and over 6X compared to the widely used Hugging Face implementation, without compromising model quality. The code is available at https://github.com/FMInference/DejaVu.