We have constructed a virtual compound library with tens of millions of compounds specifically to accelerate the drug discovery and development process. The uniqueness of this compound library lies in its construction based on synthesizability, ensuring that all included molecules can be efficiently synthesized through simple chemical reactions. By incorporating advanced machine learning techniques, the compound library can learn and predict from over a thousand known chemical reactions, thus generating a series of molecules with highly feasible synthetic routes. The distribution of parameters such as molecular weight and logP of the molecules in the library conforms to the requirements of medicinal chemistry and can be further customized according to specific needs