 Being a data scientist at Red Hat is pretty exciting because not only are you part of data science culture which involves using open-source data science tooling and looking forward to the most recent machine learning models But you're doing that at Red Hat, which is the biggest open-source company in the world. So things like contributing code upstream and focusing on the hybrid cloud and Containerizing your code is highly encouraged My name is Don Chesworth. I'm a principal data scientist here at Red Hat I've been here for about five years. My passion is deep learning machine learning neural network models and Doing that with the most cutting-edge machine learning tooling and doing that by just Distributing that work the training and the inference across multiple GPUs Right now. I have a few high torch deep learning models. Some of them are BERT sentiment models I also have a recommender system because I containerize that code and it's portable So I can read and write from S3 or Ceph or Redshift. I can have it run on my local machine I run it sometimes on a bare metal server that we have I run it on the cloud And I run it on OpenShift and a pod. I also run it with an open data hub that is also on OpenShift and My code is set up in a way that it can use a CPU if that's all that's available or a GPU or Multi-GPUs if that's possible one thing that I noticed in containerizing machine learning and distributing that across GPUs is that the way that containers are built by Default is to be nimble, obviously But because of that there's very little shared memory space on a container And you have to jump through quite a few hoops to increase that shared memory size so I worked with the open data hub team and they contributed upstream to cryo and Made a change to make it a lot easier to change your shared memory size that change went into cryo 1.20 which then went into kubernetes 1.20 and I hope if the releases line up right then in the future when OpenShift 4.7 comes out That it would be part of OpenShift 4.7. So I'm looking forward to that Everyone can look forward to having machine learning distributed across multiple GPUs be much faster much easier in OpenShift in the future and Thanks for your time