INFERENCE WITH INLINE REAL-TIME ML MODELS IN APPLICATIONS

Described are examples for using codelets executing within applications to use machine-learning (ML) models to infer a result based on application data. The codelets may be dynamically loaded into the applications during execution. A controller verifies, based on extended Berkeley packet filter (eBPF) bytecode of the codelet, that the codelet satisfies safety requirements for execution within the application. A computing device executing the application loads the verified codelet into a library of the application. The application executes the verified codelet to apply application data to the machine-learning model to infer a result. The ML model may be implemented by the eBPF code of the codelet or the codelet may include a call to a machine-learning model of a type supported by a controller of the application and a map for a serial representation of the machine-learning model. The computing device may reconstruct the ML model based on the serial representation.