Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Taxonomy of Foundation Model based Systems through the Lens of Software Architecture

Qinghua Lu, Liming Zhu, Xiwei Xu, Yue Liu, Zhenchang Xing, Jon Whittle

TL;DR

This paper addresses the lack of a systematic software-architecture approach for foundation-model-based systems. It proposes a taxonomy with three categories: foundation model pretraining/adaptation, architecture design, and responsible-AI-by-design, derived from a systematic literature review. The taxonomy clarifies design options (e.g., external vs sovereign models, retrieval-based data integration, chained versus parallel foundation models, and RLHF/guardrails) and their trade-offs in cost, accuracy, and responsible AI properties. By offering concrete guidance for architectural decisions and risk management, the work enables engineers to build safer and more effective GenAI-enabled systems and lays the groundwork for future design-pattern development.

Abstract

The recent release of large language model (LLM) based chatbots, such as ChatGPT, has attracted huge interest in foundation models. It is widely believed that foundation models will serve as the fundamental building blocks for future AI systems. As foundation models are in their early stages, the design of foundation model based systems has not yet been systematically explored. There is limited understanding about the impact of introducing foundation models in software architecture. Therefore, in this paper, we propose a taxonomy of foundation model based systems, which classifies and compares the characteristics of foundation models and design options of foundation model based systems. Our taxonomy comprises three categories: the pretraining and adaptation of foundation models, the architecture design of foundation model based systems, and responsible-AI-by-design. This taxonomy can serve as concrete guidance for making major architectural design decisions when designing foundation model based systems and highlights trade-offs arising from design decisions.

A Taxonomy of Foundation Model based Systems through the Lens of Software Architecture

TL;DR

This paper addresses the lack of a systematic software-architecture approach for foundation-model-based systems. It proposes a taxonomy with three categories: foundation model pretraining/adaptation, architecture design, and responsible-AI-by-design, derived from a systematic literature review. The taxonomy clarifies design options (e.g., external vs sovereign models, retrieval-based data integration, chained versus parallel foundation models, and RLHF/guardrails) and their trade-offs in cost, accuracy, and responsible AI properties. By offering concrete guidance for architectural decisions and risk management, the work enables engineers to build safer and more effective GenAI-enabled systems and lays the groundwork for future design-pattern development.

Abstract

The recent release of large language model (LLM) based chatbots, such as ChatGPT, has attracted huge interest in foundation models. It is widely believed that foundation models will serve as the fundamental building blocks for future AI systems. As foundation models are in their early stages, the design of foundation model based systems has not yet been systematically explored. There is limited understanding about the impact of introducing foundation models in software architecture. Therefore, in this paper, we propose a taxonomy of foundation model based systems, which classifies and compares the characteristics of foundation models and design options of foundation model based systems. Our taxonomy comprises three categories: the pretraining and adaptation of foundation models, the architecture design of foundation model based systems, and responsible-AI-by-design. This taxonomy can serve as concrete guidance for making major architectural design decisions when designing foundation model based systems and highlights trade-offs arising from design decisions.
Paper Structure (20 sections, 11 figures)

This paper contains 20 sections, 11 figures.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Methodology
  4. Design Taxonomy
  5. Foundation model pretraining and adaptation
  6. Using conventional AI models or foundation models
  7. Using external foundation models or sovereign foundation models
  8. Pretraining data
  9. Foundation model adaptation
  10. Architecture design of foundation model based systems
  11. Architectural role: foundation-model-as-a-connector
  12. Task execution with foundation models
  13. Integration of internal data
  14. Model-level architecture
  15. Deployment of foundation models
  16. ...and 5 more sections

Figures (11)

  • Figure 1: Methodology.
  • Figure 2: Taxonomy of foundation model based systems.
  • Figure 3: Conventional AI models or foundation models.
  • Figure 4: External foundation models or sovereign foundation models.
  • Figure 5: Pretraining data.
  • ...and 6 more figures