Fudan University
academicFudan built China's most storied open-LLM lineage — and then watched it migrate. Xipeng Qiu's OpenMOSS group (the MOSS lineage, SpeechGPT, the interpretability program) is now hosted at the Shanghai Innovation Institute as "OpenMOSS (SII)" ([email protected]) and tied to the MOSI.AI startup: MOSS-Speech and the 2026 omni-modal MOSS ecosystem (MOSS-TTS, MOSS-Audio, MOSS-VL, MOVA) ship SII/MOSI-branded and are tracked under SII — this entry deliberately excludes them.
What ships under Fudan branding is still substantial. The pre-SII speech line is the anchor: SpeechGPT-2.0-preview (real-time end-to-end spoken dialogue, <200ms latency), MOSS-TTSD dialogue TTS, and AnyGPT, an early any-to-any omni-modal LLM — all descendants of MOSS, one of China's first public ChatGPT-style models (Feb 2023). OpenMOSS's interpretability program (Llama Scope, 256 open SAEs on Llama-3.1-8B) began here before its follow-ons added SII affiliations.
Beyond OpenMOSS: FudanNLP (Xuanjing Huang, Tao Gui) drives agent-RL infrastructure (AgentGym-RL, with ByteDance) and the LLMEval evaluation line; FudanDISC (Zhongyu Wei) leads LLM social simulation (SocioVerse) and medical LLMs (DISC-MedLLM); Yanghua Xiao's KnowledgeWorks anchors knowledge-graph/knowledgeable-FM research; and the Fudan AI Innovation & Incubation Institute (Hao Li group) pretrains the FuXi weather/ocean forecasting FMs that beat ECMWF operational baselines.
People
- Xipeng Qiu Google Scholar — Professor, CS; leads OpenMOSS (MOSS lineage, SpeechGPT, interpretability) — recent output ships SII/MOSI-branded
- Xuanjing Huang Google Scholar — Professor, CS; long-time NLP lead (FudanNLP) — foundation models, agents, alignment, evaluation
- Tao Gui Google Scholar — Fudan NLP; agent-RL and evaluation (AgentGym-RL, AnyGPT, LLMEval)
- Yanghua Xiao Google Scholar — Professor, CS; Director, KnowledgeWorks (KW@FUDAN) — knowledge-graph / knowledgeable-foundation-model research
- Zhongyu Wei Google Scholar — Associate Professor, School of Data Science; FudanDISC social-agent / LLM social-simulation lead