Electronic device for distributed processing of artificial intelligence model and operation method of the electronic device

An electronic device configured to perform distributed processing of an (AI) model, the electronic device comprising:

• memory storing information of an AI model to be distributed and processed;
• at least one processor configured to:
• divide the AI model to obtain a plurality of submodels processable in parallel and having respective computational requirements,
• determine a first processing device from among a plurality of processing device to process a first submodel requiring a highest computational requirements from among the respective computational requirements of the submodels,
• predict a first processing time required for the first processing device to process the first submodel, and
• determine a second processing device from among the processing devices to process a second submodel from among the submodels based on the first predicted processing time;
• and
• a communication unit comprising circuitry configured to transmit, to the first processing device and the second processing device, information for requesting to process the first submodel and the second submodel, respectively,
• wherein the at least one processor is configured to:
• determine guard time information based on adding a value to a predicted processing time required to process each submodel of at least one submodel among the plurality of the submodels;
• identify a submodel for which a result of processing the at least one submodel is not output until a preset time point, based on the guard time information;
• and
• determine a processing device to process the identified submodel, and request the determined processing device to process the identified submodel.

A non-transitory computer-readable recording medium having recorded thereon a program which, when executed by at least one processor, causes the at least one processor to perform:

• obtaining an artificial intelligence (AI) model to be distributed and processed;
• dividing the AI model to obtain a plurality of submodels processable in parallel and having respective computational requirements;
• determining a first processing device from among a plurality of processing devices to process a first submodel requiring a highest computational requirements from among the respective computational requirements of the submodels;
• predicting a first processing time required for the first processing device to process the first submodel;
• determining a second processing device from among the processing devices to process a second submodel from among the submodels based on the predicted first processing time;
• performing distributed processing of the AI model by requesting the first processing device and the second processing device to process the first submodel and the second submodel, respectively;
• determining guard time information based on adding a value to a predicted processing time required to process each submodel of at least one submodel among the plurality of the submodels;
• identifying a submodel for which a result of processing the at least one submodel is not output until a preset time point, based on the guard time information;
• and
• determining a processing device to process the identified submodel, and requesting the determined processing device to process the identified submodel.