Chameleon: a Heterogeneous and Disaggregated Accelerator System for Retrieval-Augmented Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiang, Wenqi, Zeller, Marco, Waleffe, Roger, Hoefler, Torsten, Alonso, Gustavo
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909550432485376
author Jiang, Wenqi
Zeller, Marco
Waleffe, Roger
Hoefler, Torsten
Alonso, Gustavo
author_facet Jiang, Wenqi
Zeller, Marco
Waleffe, Roger
Hoefler, Torsten
Alonso, Gustavo
contents A Retrieval-Augmented Language Model (RALM) combines a large language model (LLM) with a vector database to retrieve context-specific knowledge during text generation. This strategy facilitates impressive generation quality even with smaller models, thus reducing computational demands by orders of magnitude. To serve RALMs efficiently and flexibly, we propose Chameleon, a heterogeneous accelerator system integrating both LLM and vector search accelerators in a disaggregated architecture. The heterogeneity ensures efficient serving for both inference and retrieval, while the disaggregation allows independent scaling of LLM and vector search accelerators to fulfill diverse RALM requirements. Our Chameleon prototype implements vector search accelerators on FPGAs and assigns LLM inference to GPUs, with CPUs as cluster coordinators. Evaluated on various RALMs, Chameleon exhibits up to 2.16$\times$ reduction in latency and 3.18x speedup in throughput compared to the hybrid CPU-GPU architecture. The promising results pave the way for adopting heterogeneous accelerators for not only LLM inference but also vector search in future RALM systems.
format Preprint
id arxiv_https___arxiv_org_abs_2310_09949
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Chameleon: a Heterogeneous and Disaggregated Accelerator System for Retrieval-Augmented Language Models
Jiang, Wenqi
Zeller, Marco
Waleffe, Roger
Hoefler, Torsten
Alonso, Gustavo
Machine Learning
Artificial Intelligence
Hardware Architecture
Computation and Language
A Retrieval-Augmented Language Model (RALM) combines a large language model (LLM) with a vector database to retrieve context-specific knowledge during text generation. This strategy facilitates impressive generation quality even with smaller models, thus reducing computational demands by orders of magnitude. To serve RALMs efficiently and flexibly, we propose Chameleon, a heterogeneous accelerator system integrating both LLM and vector search accelerators in a disaggregated architecture. The heterogeneity ensures efficient serving for both inference and retrieval, while the disaggregation allows independent scaling of LLM and vector search accelerators to fulfill diverse RALM requirements. Our Chameleon prototype implements vector search accelerators on FPGAs and assigns LLM inference to GPUs, with CPUs as cluster coordinators. Evaluated on various RALMs, Chameleon exhibits up to 2.16$\times$ reduction in latency and 3.18x speedup in throughput compared to the hybrid CPU-GPU architecture. The promising results pave the way for adopting heterogeneous accelerators for not only LLM inference but also vector search in future RALM systems.
title Chameleon: a Heterogeneous and Disaggregated Accelerator System for Retrieval-Augmented Language Models
topic Machine Learning
Artificial Intelligence
Hardware Architecture
Computation and Language
url https://arxiv.org/abs/2310.09949