Shanghai AI lab launches open-source multimodal model with advanced memory and scientific task capabilities

Shanghai AI Laboratory has released the foundational large model Shusen-S2, which demonstrates top-tier performance in various scientific domains. The model features a Memory Decoder architecture with a plug-and-play memory module, allowing for domain-specific expertise to be added without retraining the entire model. This innovation enables the model to maintain its general capabilities while expanding into specialized fields. In addition to its memory mechanism, Shusen-S2 enhances long-range reasoning and agent execution capabilities, crucial for complex scientific research tasks. These features are designed to support multi-step problem-solving and tool usage, positioning the model as a serious contender in scientific research. The release marks a significant step toward making open-source models capable of handling advanced scientific challenges.