Resource usage prediction for deep learning model
According to implementations of the subject matter described herein, there is provided a solution for predicting the resource usage of the deep learning model. In this solution, information about a deep learning model is obtained, the information comprising first information for describing the deep learning model and second information about an operating environment of a job associated with the deep learning model. The static resource usage of the job is determined based on the first information and a strategy of the job during runtime in the operating environment is determined. Afterwards, resource usage of the job during runtime in the operating environment is predicted based on the strategy and the static resource usage. With this solution, the usage of various resources of the deep learning model, such as computation power consumption, memory consumption, execution time, and the like, under a specific runtime strategy can be accurately predicted.
1 . A computer-implemented method, comprising: obtaining information about a deep learning model, the information comprising first information for describing the deep learning model and second information about an operating environment of a job associated with the deep learning model; wherein the first information comprises a configuration parameter of the deep learning model and second information comprises at least one of: a framework type of the deep learning model, specifications and number of computing devices for executing the job in the operating environment, and an execution strategy of the job on the computing devices; determining static resource usage of the job based on the first information; wherein determining the static resource usage comprises: generating, based on the first information, a computation graph corresponding to the deep learning model, the computation graph comprising a plurality of nodes corresponding to a plurality of operators in the deep learning model and edges connecting the plurality of nodes indicating dependency among the plurality of operators: predicting, based on the computation graph and respective resource prediction models of the plurality of operators, respective static resource usage of the plurality of operators; and determining the static resource usage of the job based on the respective static resource usage of the plurality of operators: determining, based on the first information and the second information, a strategy of the job during runtime in the operating environment; and predicting, based on the strategy and the static resource usage, resource usage of the job during runtime in the operating environment.
2 . The method of claim 1 , wherein the first information comprises at least one of: a model file of the deep learning model and program codes of the job.
3 . The method of claim 1 , wherein the resource usage comprises at least one of: computation power consumption, memory consumption, I/O resource consumption, execution time, and power consumption.
4 . The method of claim 3 , wherein the resource usage comprises other resource consumption determined based on at least one of the computation power consumption and the memory consumption.
5 . The method of claim 1 , wherein the strategy comprises at least one of: a resource allocation strategy of the deep learning model, and an execution strategy of the job in the operating environment.
6 . The method of claim 5 , wherein predicting the resource usage of the job during runtime in the operating environment comprises: adjusting the static resource usage based on at least one of the resource allocation strategy and the execution strategy, to obtain the resource usage of the job during runtime in the operating environment.
7 . The method of claim 1 , further comprising: generating, using a trained machine learning model, a parameter for optimizing the predicted resource usage; and optimizing, based on the parameter, the predicted resource usage.
8 . An electronic device, comprising: a processing unit; and a memory coupled to the processing unit and comprising instructions stored thereon which, when executed by the processing unit, cause the device to perform acts comprising: obtaining information about a deep learning model, the information comprising first information for describing the deep learning model and second information about an operating environment of a job associated with the deep learning model; wherein the first information comprises a configuration parameter of the deep learning model and second information comprises at least one of: a framework type of the deep learning model, specifications and number of computing devices for executing the job in the operating environment, and an execution strategy of the job on the computing devices; determining static resource usage of the job based on the first information; wherein determining the static resource usage comprises: generating, based on the first information, a computation graph corresponding to the deep learning model, the computation graph comprising a plurality of nodes corresponding to a plurality of operators in the deep learning model and edges connecting the plurality of nodes indicating dependency among the plurality of operators; predicting, based on the computation graph and respective resource prediction models of the plurality of operators, respective static resource usage of the plurality of operators; and determining the static resource usage of the job based on the respective static resource usage of the plurality of operators; determining, based on the first information and the second information, a strategy of the job during runtime in the operating environment; and predicting, based on the strategy and the static resource usage, resource usage of the job during runtime in the operating environment.
9 . The device of claim 8 , wherein the resource usage comprises at least one of: computation power consumption, memory consumption, I/O resource consumption, execution time, and power consumption.
10 . The device of claim 9 , wherein the resource usage comprises other resource consumption determined based on at least one of the computation power consumption and the memory consumption.
11 . The device of claim 8 , wherein determining the static resource usage comprises: generating, based on the first information, a computation graph corresponding to the deep learning model, the computation graph comprising a plurality of nodes corresponding to a plurality of operators in the deep learning model and edges connecting the plurality of nodes indicating dependency among the plurality of operators; predicting, based on the computation graph and respective resource prediction models of the plurality of operators, respective static resource usage of the plurality of operators; and determining the static resource usage of the job based on the respective static resource usage of the plurality of operators.
12 . A computer program product being tangibly stored in a non-transitory computer storage medium and comprising machine-executable instructions which, when executed by a device, cause the device to perform acts comprising: obtaining information about a deep learning model, the information comprising first information for describing the deep learning model and second information about an operating environment of a job associated with the deep learning model; wherein the first information comprises a configuration parameter of the deep learning model and second information comprises at least one of: a framework type of the deep learning model, specifications and number of computing devices for executing the job in the operating environment, and an execution strategy of the job on the computing devices: determining static resource usage of the job based on the first information; wherein determining the static resource usage comprises: generating, based on the first information, a computation graph corresponding to the deep learning model, the computation graph comprising a plurality of nodes corresponding to a plurality of operators in the deep learning model and edges connecting the plurality of nodes indicating dependency among the plurality of operators: predicting, based on the computation graph and respective resource prediction models of the plurality of operators, respective static resource usage of the plurality of operators; and determining the static resource usage of the job based on the respective static resource usage of the plurality of operators; determining, based on the first information and the second information, a strategy of the job during runtime in the operating environment; and predicting, based on the strategy and the static resource usage, resource usage of the job during runtime in the operating environment.