Abstract: An example system includes prediction workers, training workers, and a parameter server. The prediction workers store a local copy of a machine-learned model and run the mode exclusively in serving mode. The training workers store a local copy of a machine-learned model and a local snapshot and run the local copy exclusively in training mode and compare the local model or state to the snapshot after training to send delta updates to the parameter server after training. The parameter server aggregates received delta updates into a master copy of the model, sends the aggregated updates back to training workers and provides two types of updates; a real-time update based on a comparison of the master model with a local snapshot, and a full update. The real-time update occurs at least an order of magnitude more frequently than the full update and includes a subset of the weights in the model.

Abstract: To present one or more content to users of an online system, the online system identifies a content evaluation pipeline including an order of a plurality of stages having one or more computer models for evaluating a likelihood of user interaction with a content item. The content evaluation pipeline selects a decreasing number of content items, from each stage of the order, according to the order of the stages in the order. The online system identifies a set of candidate modifications to one or more operational parameters of the content evaluation pipeline. For each candidate modification, the online system determines a compute time value and a content selection value. For a given amount of compute time, the online system optimizes the one or more operational parameters based on the determined content time value and the determined content selection value to increase the content selection value of the content evaluation pipeline.

Abstract: To present one or more content to users of an online system, the online system identifies a content evaluation pipeline including an order of a plurality of stages having one or more computer models for evaluating a likelihood of user interaction with a content item. The content evaluation pipeline selects a decreasing number of content items, from each stage of the order, according to the order of the stages in the order. The online system identifies a set of candidate modifications to one or more operational parameters of the content evaluation pipeline. For each candidate modification, the online system determines a compute time value and a content selection value. For a given amount of compute time, the online system optimizes the one or more operational parameters based on the determined content time value and the determined content selection value to increase the content selection value of the content evaluation pipeline.