m-a-p/YuE2-3B

YuE2-3B is an open music generation model that converts lyrics and style prompts into songs with vocals. It uses symbolic planning and an AR-NAR Mixture-of-Transformers backbone. Developers can run it locally on a 24GB GPU without quantization. The model card provides installation steps for the inference package and documents APIs for text guidance and agentic editing of scores.