4 ms·
Anything that promise k8s the easy way is a click-bait, I'm a ML researcher and was promised that KubeFlow will help me deploy my models from my notebook with t
by spicyramen 7y ago
Anything that promise k8s the easy way is a click-bait, I'm a ML researcher and was promised that KubeFlow will help me deploy my models from my notebook with the infrastructure I needed...well it happens to be that we didn't have Kubernetes. I started with KubeFlow then K8s and then I got lost in all the terminology, which became a full time job, not as ML researcher but as DevOps. K8s is powerful but for my use case is a non-starter: maintain all the components+infra and don't talk about day 2 operations because I would have no idea where to start if something fails. I will definitely prefer Sagemaker or Cloud ML engine, and if get the budget maybe hire a k8s expert to handle that
- calebkaiser 7y agoShameless self-plug, but have you looked at Cortex? It deploys trained models as APIs and automates the underlying devops. It's also free and open source. GitHub: https://github.com/cortexlabs/cortex https://github.com/cortexlabs/cortex
- spicyramen 7y agoThanks this looks promising will take a look
- StreamBright 7y agoMinor clarification, knowing K8s is not DevOps. You were boarding a hype train that is fueled by many parties and just needed to realize that it is going in the wrong direction. Luckily you got off on time there are higher abstractions for ML as you pointed out, which you can use with less time investment.
- spicyramen 7y agoThanks, k8s would be Infrastructure I guess?