Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time Article Swipe

MAKE IMPACT

Zichang Liu , Jue Wang , Tri Dao , Tianyi Zhou , Binhang Yuan , Zhao Song , Anshumali Shrivastava , Ce Zhang , Yuandong Tian , Christopher Ré , Beidi Chen · YOU? · · 2023 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2310.17157

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.

Type

preprint

Language

en

Landing Page

http://arxiv.org/abs/2310.17157

PDF

https://arxiv.org/pdf/2310.17157

OA Status

green

Cited By

19

Related Works

10

OpenAlex ID

https://openalex.org/W4387995158

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W4387995158

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.48550/arxiv.2310.17157

Digital Object Identifier

Title

Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time

Work title

Type

preprint

OpenAlex work type

Language

en

Primary language

Publication year

2023

Year of publication

Publication date

2023-10-26

Full publication date if available

Authors

Zichang Liu, Jue Wang, Tri Dao, Tianyi Zhou, Binhang Yuan, Zhao Song, Anshumali Shrivastava, Ce Zhang, Yuandong Tian, Christopher Ré, Beidi Chen

List of authors in order

Landing page

https://arxiv.org/abs/2310.17157

Publisher landing page

PDF URL

https://arxiv.org/pdf/2310.17157

Direct link to full text PDF

Open access

Yes

Whether a free full text is available

OA status

green

Open access status per OpenAlex

OA URL

https://arxiv.org/pdf/2310.17157

Direct OA link when available

Concepts

Computer science, Inference, Speedup, Asynchronous communication, Context (archaeology), Exploit, Machine learning, Artificial intelligence, Latency (audio), Deep learning, Parallel computing, Computer security, Biology, Telecommunications, Computer network, Paleontology

Top concepts (fields/topics) attached by OpenAlex

Cited by

19

Total citation count in OpenAlex

Citations by year (recent)

2025: 7, 2024: 10, 2023: 2

Per-year citation counts (last 5 years)

Related works (count)

10

Other works algorithmically related by OpenAlex

Full payload