16:30 - 17:15
Applying machine learning in online applications requires solving the problem of model serving: Evaluating the machine-learned model over some data point(s) in real time while the user is waiting for a response. Solutions such as TensorFlow Serving are available to solve this problem where the model only needs to be evaluated over a one data point per user request, but this is not sufficient for problems where many data points must be evaluated to make a decision, such as in search and recommendation. This talk will show that this is a bandwidth constrained problem, and outline an architectural solution where computation is pushed down to data shards in parallel. It will demonstrate how this solution can be put into use with Vespa.ai, an open source engine, to achieve scalable model serving of TensorFlow and ONNX, and show benchmarks comparing performance and scalability to TensorFlow Serving. Model serving with Vespa is used today for some of the world’s largest recommender systems, such as serving personalized content on all Yahoo content pages and personalized ads in the world’s third-largest ad network. These systems evaluate models over millions of data points per request for hundreds of thousands of requests per second.