This is based upon this repo In this blog post we will dig down in to how a Machine Learning(ML) model can be combined with pre and post processing steps using Nvidia triton. By combining the pre- and post processing the user can make a single call using GPRC or http. It should be noted that we in reality will not merge these processing steps in any way but link the calls together using Tritons ensemble functionality Triton support multiple different backends(processing functionality) and in this case we will use the tensorRT backend for the model serving and the python backend to add the pre and post processing business logic.…

“Yeah It’s on. ” Hello World! …

Poor mans datalake layout: post title: “Poor mans datalake” subtitle: “DuckDb” date: 2023-02-01 author: “Niklas Hansson” URL: “/2023/02/01/Trition_with_post_and_pre_processing/” This post is a deep dive playing with DuckDB doing a twist on Build a poor man’s data lake from scratch with DuckDB where we will do the following changes: Use Minio instead of S3 DBT instead of dagster. We will host it on Kubernetes and set it up so it all run locally.…