Shanghai AI Lab officially released the world's largest open-source parameter-scale scientific multimodal model, "ShuRen Intern-S1-Pro" on February 4. This groundbreaking model is based on the SAGE "integration of general and specialized" technology architecture, with as many as 1 trillion parameters, becoming a shining star in the current open-source community.
The core scientific capabilities of the "ShuRen Intern-S1-Pro" model have reached international leading levels, especially in high-difficulty interdisciplinary evaluations, demonstrating strong logical reasoning capabilities, even reaching the level of gold medalists in Olympic competitions. At the same time, the model ranks among the top open-source models in terms of intelligent agent capabilities in real research processes, providing researchers with a more powerful tool.
This model adopts a mixture-of-experts (MoE) architecture, with a total of 512 experts, and only 8 experts are activated each time, using 2.2 billion parameters. This design not only optimizes the model's computational efficiency but also greatly reduces resource consumption. On the underlying architecture, "ShuRen Intern-S1-Pro" has achieved two important breakthroughs. First, by introducing Fourier position encoding and reconstructing the sequence encoder, the model has obtained the "physical intuition" from microscopic life signals to macroscopic cosmic fluctuations, enhancing its understanding ability. Second, an efficient routing mechanism was adopted to solve the stability and computing power efficiency bottlenecks when training a trillion-parameter model, laying the foundation for training ultra-large-scale models.
Notably, "ShuRen Intern-S1-Pro" is not only an academic model, but also laid a solid foundation for building an open and shared AGI4S (Artificial Intelligence for Science) infrastructure in the future. Through original model architecture and self-developed computing power technology, Shanghai AI Lab demonstrated China's strength and potential in the field of artificial intelligence.
To allow more users to experience this advanced model, Shanghai AI Lab also provides online experience and open-source addresses, making it convenient for developers and researchers to explore further.
