MOSS

Overview

MOSS is a conversational large language model developed by a team at Fudan University, and it is the first publicly released model of its kind in China. In February this year, the team invited the public to join its internal testing, offering a rare opportunity for users to interact with a state-of-the-art conversational AI system.

Capable of performing a wide range of tasks—including dialogue generation, programming assistance, and factual question answering—MOSS has opened up the complete technical pathway that enables generative language models to interpret human intentions and engage in coherent, natural conversations. From pretraining and fine-tuning to alignment and dialogue optimization, the model demonstrates a full-stack solution for building conversational AI.

On April 21, the newly upgraded MOSS model was officially launched, and its source code was made open source, marking an important milestone for the open-source AI ecosystem and paving the way for further research and application development.

Core capabilities

Conversational dialogue generation

Engages in multi-turn, coherent natural conversations with users

Programming assistance

Helps generate code snippets, debug tips, and technical solutions through conversation

Factual question answering

Provides answers to factual queries based on its trained knowledge

Full-stack technical pathway

Demonstrates complete pipeline from pretraining and fine-tuning to alignment and dialogue optimization

Open-source availability

Source code and model weights are publicly released for research and development

Local deployment support

Offers deployment documentation and inference scripts for self-hosted usage

Fine-tuning capability

Allows developers to adapt the model for specific tasks or domains using open-source code

Use cases

Intelligent customer service: Deploy MOSS as an online support bot to answer common user questions

Educational tutoring: Students ask MOSS math or writing questions and receive answers and advice

Content creation: Self-media creators use MOSS to generate article outlines, headlines, or social media copy

Code assistance: Programmers quickly get code snippets or debugging tips through conversation

Knowledge sharing: Q&A communities build knowledge bases with MOSS to auto-reply to professional questions

Research experiments: Labs fine-tune MOSS and compare performance for academic studies

Enterprise training: Create an internal Q&A system to support employee skill development