OceanInstruct 海洋大模型指令数据集

OceanInstruct 是一个专为海洋科学领域设计的大型语言模型指令数据集，包含 20,000 条指令，旨在为海洋领域的大型语言模型提供训练数据。这些指令覆盖了广泛的海洋科学知识，确保模型在海洋科学问答、内容生成和水下具身智能能力等方面具备专业能力。该数据集被用于训练 OceanGPT 模型，该模型在海洋科学问答、内容生成等方面表现出色。 OceanGPT 模型在多个任务上优于基线语言模型，显示出其在处理需要专业知识的海洋任务上的优势。

OceanInstruct is a large language model instruction dataset tailored for the marine science domain, consisting of 20,000 instruction samples, which is designed to provide training data for large language models specialized in the marine field. These instructions cover a broad spectrum of marine science knowledge, enabling the model to acquire professional capabilities in marine science question answering, content generation, and underwater embodied intelligence-related tasks. This dataset is utilized to train the OceanGPT model, which delivers outstanding performance in marine science question answering, content generation, and other relevant scenarios. The OceanGPT model outperforms baseline language models across multiple tasks, highlighting its superior performance in handling professional knowledge-intensive marine tasks.

搜集汇总

背景与挑战

背景概述

OceanInstruct是浙江大学于2024年开源的海洋科学大型语言模型指令数据集，包含20,000条指令，旨在训练OceanGPT模型以提升其在海洋科学问答、内容生成等专业任务上的能力。该数据集与OceanBench基准测试一同推出，用于评估模型在海洋学领域的表现。