Fengshenbang-LM
An open-source system of Chinese pretrained language and multimodal models, led by IDEA Research Institute, spanning general, generation, and domain-specific model families.
Fengshenbang-LM is an open-source large-model program built around Chinese-language pretrained models, developed by IDEA Research Institute's Cognitive Computing and Natural Language Research Center. It provides several model families at different scales: Jiangziya, general-purpose, used for translation, coding, classification, and question answering; Erlangshen, focused on language understanding; Wenzhong, for language generation; Randeng, for text-to-text tasks such as translation and summarization; Taiyi, for cross-modal tasks including text-to-image generation and protein structure prediction; and Yuyuan, tuned for medical, financial, and legal domains.
The project also provides the Fengshen framework, training and fine-tuning scripts, model download links, and documentation, and is built on PyTorch and Hugging Face Transformers.