| CPC G06F 8/63 (2013.01) [G06F 8/10 (2013.01)] | 10 Claims |
|
1. An apparatus for generating an executable image of a neural network, comprising:
one or more processors; and
executable memory for storing at least one program executed by the one or more processors,
wherein the at least one program is configured to:
receive user requirements including a default neural network model and training result data for generating a neural network executable image required by a user;
check whether the default neural network model included in the user requirements is capable of being supported in a target system in which the neural network executable image is to be installed;
convert the default neural network model into a neural network model that is executable in the target system;
convert the training result data by reconfiguring a data format set of the training result data;
generate a neural network executable image by combining the converted neural network model and the converted training result data;
wherein the at least one program is further configured to:
check whether a neural network model conversion plugin that supports a neural network graph representation format of the default neural network model and a data format of the training result data is present, in order to check whether the default neural network model is capable of being supported in the target system;
check whether an accelerator type included in the user requirements is present in an accelerator library and identifies execution functions used by an inference engine included in the user requirements, thereby checking whether the default neural network model is capable of being supported in the target system;
convert the default neural network model comprises selecting a pivot model for neural network model conversion;
form a neural network model conversion plugin in which a function for two-way conversion between the selected pivot model and each neural network model is implemented for each neural network model; and
converts the default neural network model by checking whether the selected pivot model is a same type of model as a neural network model for inference engine in the target system.
|