Octopus v3: Technical Report for On-device Sub-billion Multimodal AI Agent

TL;DR

This paper introduces Octopus v3, a compact multimodal AI model under 1 billion parameters designed for efficient on-device processing of visual, language, and audio data, suitable for edge devices like Raspberry Pi.

Contribution

The paper presents a novel multimodal model with functional tokens, optimized for edge devices, capable of processing multiple data modalities in English and Chinese.

Findings

Operates efficiently on devices as constrained as Raspberry Pi

Supports multilingual processing in English and Chinese

Maintains high performance with less than 1B parameters

Abstract

A multimodal AI agent is characterized by its ability to process and learn from various types of data, including natural language, visual, and audio inputs, to inform its actions. Despite advancements in large language models that incorporate visual data, such as GPT-4V, effectively translating image-based data into actionable outcomes for AI agents continues to be challenging. In this paper, we introduce a multimodal model that incorporates the concept of functional token specifically designed for AI agent applications. To ensure compatibility with edge devices, our model is optimized to a compact size of less than 1B parameters. Like GPT-4, our model can process both English and Chinese. We demonstrate that this model is capable of operating efficiently on a wide range of edge devices, including as constrained as a Raspberry Pi.