FedLLM-Bench
收藏资源简介:
FedLLM-Bench是由上海交通大学和上海人工智能实验室等机构联合开发的数据集,包含四个子数据集:Fed-Aya、Fed-ChatbotIT、Fed-WildChat和Fed-ChatbotPA。这些数据集涵盖了从38到747个客户端,涉及多语言、质量、数量、指令、长度、嵌入和偏好等多个维度,旨在模拟真实世界的多语言协作场景。数据集的创建过程考虑了客户端数据的自然分割,确保了数据的真实性和多样性。FedLLM-Bench的应用领域主要集中在联邦学习大型语言模型的性能评估和方法比较,以及推动新研究方向的探索。
FedLLM-Bench is a dataset jointly developed by Shanghai Jiao Tong University, Shanghai AI Laboratory and other institutions. It comprises four sub-datasets: Fed-Aya, Fed-ChatbotIT, Fed-WildChat and Fed-ChatbotPA. These datasets cover client scales ranging from 38 to 747, and involve multiple dimensions including multilingualism, data quality, data volume, instructions, sequence length, embeddings and preferences, aiming to simulate real-world multilingual collaborative scenarios. The dataset creation process considers the natural partitioning of client-side data, ensuring the authenticity and diversity of the collected data. The application fields of FedLLM-Bench mainly focus on performance evaluation and method comparison for federated learning large language models (LLMs), as well as promoting the exploration of new research directions.




