Shanghai Jiaotong University released the open source Magnolia science model, which includes three fields: chemical synthesis, neural fluid and scientific evaluation benchmark.

Thanks to CTOnews.com netizens Yuan Sikai for the delivery of the clue! CTOnews.com, July 12, according to the News academic Network of Shanghai Jiaotong University, the university first launched Magnolia science model 2023 at the just-concluded World artificial Intelligence Congress.

According to reports, the Magnolia science model consists of three domain models, which are released by Shanghai Magnolia Open Source Research Institute:

Chemical synthesis (BAI-Chem) represents the direction of the discipline centered on molecular structure.

Neural fluid (BAI-NeuroFluid) representing the direction of the discipline centered on dynamics

Benchmark for scientific evaluation of interdisciplinary and cross-modal basic models (BAI-SciEval)

Among them, Magnolia scientific model-chemical synthesis claims to be the first large chemical model with the ability of reaction condition generation and "human in the loop" feedback optimization, which can accelerate the process of molecular design and reaction design, as well as the process of chemical synthesis of full chains, potential functional molecules (such as drug molecules) and their synthesis schemes, such as condition generation and reaction testing. It can speed up the overall time "from a few months in the traditional method to dozens of minutes".

The platform also provides chemical chat robot and chemical synthesis assistant, which supports users to interact in natural language, and can provide generative synthesis scheme and retrieval synthesis scheme at one time based on the target molecule given by the user.

Magnolia science model-neural fluid claims to have created a visual intuitive learning framework. Shanghai Jiaotong University said that the model can apply physically inspired generative AI technology to the calculation and simulation of fluid particle models, and "realize internal fluid dynamics from external visual apparent reasoning for the first time". In the inversion of fluid properties such as viscosity and density, in the aspects of problem scale, simulation speed, model generalization, inverse problem solving accuracy, etc. It also "breaks through the calculation bottleneck of the existing fluid numerical simulation methods".

Magnolia magnolia science model-evaluation benchmark is the first evaluation benchmark with dynamic data built for the field of basic science, which claims to cover about 18000 challenging scientific issues. it involves the three basic science fields of chemistry, physics and biology and their dozens of seed fields, including basic knowledge, knowledge application, scientific computing and scientific research ability. The purpose of this paper is to comprehensively evaluate the scientific knowledge mastery level and auxiliary scientific research ability of the large model.

CTOnews.com includes links to Magnolia open source official website and official GitHub page:

Baiyulan Open AI

GitHub of Shanghai Magnolia Open Source Research Institute