OceanGPT

沧渊-Ocean Foundation Models

Introduction

The Cangyuan Ocean Model (OceanGPT) was jointly developed by a multidisciplinary team organized by the National Key Laboratory of Ocean Precision Perception (Zhejiang University). In recent years, the rapid development of general artificial intelligence represented by big language models has brought new opportunities for marine technology. Cangyuan is committed to exploring a new path of universal AI technology with ocean characteristics based on new AI technologies such as big language models, knowledge graphs, generative AI, and embodied agents. The development of "Cangyuan" began in 2023, with the first open-source big language model in the ocean field, OceanGPT-1, being released. Subsequently, OceanGPT basic (domain knowledge enhanced ocean knowledge question answering big model), OceanGPT omni (ocean multimodal big model), and OceanGPT coder (underwater embodied robot code big model) were continuously updated and released.

Welcome

Marine Applications

Marine Professional Q&A

Understanding multimodal ocean data

Underwater robot control and underwater embodied intelligence

OceanGPT Model System

OceanGPT adopts a multi expert hybrid architecture, consisting of multiple functional sub models working together. OceanGPT basic is responsible for understanding and answering questions related to ocean basic professional knowledge. OceanGPT o is positioned for natural language interpretation of ocean specific multimodal perception data such as sonar, optics, and electromagnetics. OceanGPT encoder supports the generation of control instructions for underwater robots. At the same time, OceanGPT can integrate general models such as DeepSeek, and integrate general language understanding and marine professional abilities through a "combination of general and specialized" mode.