目录:
The OceanGPT Large Model (沧渊, "Cangyuan") was jointly developed by a multidisciplinary team organized by the State Key Laboratory of Ocean Precision Sensing at Zhejiang University. In recent years, the rapid advancement of general artificial intelligence, particularly large language models (LLMs), has brought new opportunities to marine science and technology. OceanGPT is dedicated to exploring new paths for general AI technologies with marine-specific characteristics, leveraging cutting-edge AI techniques such as large language models, knowledge graphs, generative AI, and embodied agents.
The development of OceanGPT began in 2023, with the release of OceanGPT-1, the first open-source large language model tailored for the marine domain. Subsequent versions have been continuously released, including:
OceanGPT adopts a Mixture-of-Experts (MoE) architecture, composed of multiple specialized sub-models working collaboratively:
In addition, OceanGPT can be integrated with general-purpose models such as DeepSeek, enabling a hybrid approach that combines general language understanding with domain-specific expertise in ocean sciences.
OceanGPT can be applied across a wide range of scenarios in the marine domain, including marine question answering, multimodal understanding of oceanographic data, underwater robotic control, and embodied intelligence in underwater environments. It provides foundational support for a variety of marine applications such as underwater archaeology, seafloor exploration, oceanographic research, and monitoring of submarine cables and pipelines.