Timezone: »

Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time
Zichang Liu · Jue Wang · Tri Dao · Tianyi Zhou · Binhang Yuan · Zhao Song · Anshumali Shrivastava · Ce Zhang · Yuandong Tian · Christopher Re · Beidi Chen

Thu Jul 27 06:48 PM -- 06:56 PM (PDT) @ Ballroom A
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 2$\times$ compared to the state-of-the-art FasterTransformer, and over 6$\times$ compared to the widely used Hugging Face implementation, without compromising model quality. The code is available at https://github.com/FMInference/DejaVu.

Author Information

Zichang Liu (Rice University)
Jue Wang (ETH Zürich)
Tri Dao (Stanford)
Tianyi Zhou
Binhang Yuan (Swiss Federal Institute of Technology)
Zhao Song (Adobe Research)
Anshumali Shrivastava (Rice University)

Anshumali Shrivastava is an associate professor in the computer science department at Rice University. His broad research interests include randomized algorithms for large-scale machine learning. In 2018, Science news named him one of the Top-10 scientists under 40 to watch. He is a recipient of National Science Foundation CAREER Award, a Young Investigator Award from Air Force Office of Scientific Research, and machine learning research award from Amazon. His research on hashing inner products has won Best Paper Award at NIPS 2014 while his work on representing graphs got the Best Paper Award at IEEE/ACM ASONAM 2014. Anshumali finished his Ph.D. in 2015 from Cornell University.

Ce Zhang (ETH Zurich)
Yuandong Tian (Facebook AI Research)
Christopher Re (Stanford University)
Beidi Chen (CMU / FAIR)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors