| CPC G06F 16/24556 (2019.01) [G06F 21/6218 (2013.01)] | 18 Claims |
|
1. A method for managing computer model repositories, the method performed by a server comprising a computer processor, the method comprising:
aggregating a plurality of repositories, each of the plurality of repositories comprising one or more model data elements of a machine learning model, wherein the plurality of repositories include: a remote repository, and a local repository, wherein the remote repository is comprised in a first computer system, the first computer system geographically separated from the server;
wherein the aggregating of the plurality of repositories comprises:
proxying, in the local repository, one or more model data elements from the remote repository, wherein the proxying comprises strategically storing, at the server, one or more of the model data elements from the remote repository in a cache; and
associating a single address with: the remote repository, and the local repository;
in response to a command comprising the single address, the command received from a second computer system, the second computer system geographically separated from the server:
determining a state for the machine learning model, wherein the determining of the state comprises comparing a version of the machine learning model stored in the local repository to a reference version of the machine learning model stored in the remote repository, the comparing using at least one of: a checksum operation, a model commit, and a hash function, wherein the version of the machine learning model stored in the local repository comprises one or more of the proxied model data elements; and
fetching one or more of the model data elements based on the determined state, wherein the fetching of one or more of the model data comprises transmitting one or more of the model data elements to the second computer system;
wherein the command received from the second computer system is a request for a version of the machine learning model made to the single address, wherein the server checks for the requested version in the remote repository, and
for a requested version existing in the remote repository:
validating the requested version existing in the remote repository; and
retrieving one or more model data elements of the requested version from the remote repository and storing the one or more retrieved model data elements in the cache;
wherein the fetching of the one or more model data elements comprises transmitting one or more of the model data elements retrieved from the remote repository and stored in the cache to the second computer system.
|