120 points by datasciencedojo 6 months ago flag hide 30 comments
jamesmsick 6 months ago next
Kubeflow is a fantastic tool for automated ML model training and deployment! Really looking forward to seeing more projects use this in production.
turingtested 6 months ago prev next
I've been using Kubeflow for a while now and it's amazing how much time it saves during the development cycle. Anyone else on HN using this?
alexis_mancin 6 months ago next
I use Kubeflow too. I think it's a game changer for ML model deployment. I'd love to hear more about your use case though.
alexis_mancin 6 months ago next
I'm currently working on a project that incorporates both data and feature pipelines. I've heard that Kubeflow handles such workloads really well. Thoughts?
mshan 6 months ago prev next
As a reference, I've put together a simple guide that compares Kubeflow to other ML platforms like TensorFlow Serving and SageMaker. Happy to share it if there's interest!
msullivan01 6 months ago prev next
How does Kubeflow's performance compare to other ML platforms like TensorFlow Serving?
hankchizljaw 6 months ago prev next
Kubeflow makes it really easy to move through different development environments. Has anyone tried using it for cross-cloud deployment?
jchen01 6 months ago next
Yes, we have been using Kubeflow for multi-cloud deployment. It provides a unified development experience and reduces deployment time significantly.
strider 6 months ago prev next
Here is a post describing a real-world multi-cloud deployment case study with Kubeflow: <https://medium.com/swlh/kubeflow-multi-cloud-deployment-case-study-309cf58e08da>
briandoll 6 months ago prev next
Hi James, can I get your thoughts on some of the challenges faced when deploying ML models in production? Happy to discuss here or DM.
bigprendy 6 months ago prev next
I'd love to see a performance comparison between Kubeflow and other cloud ML services for model training. Has anyone here tried it?
mitchellk999 6 months ago prev next
I'm excited to see Kubeflow's development and adoption within the ML community. The recent addition of Argo truly makes Kubeflow a full CI/CD solution for ML.
randomcoder 6 months ago next
@mitchellk999 I believe you're right - I've found it to be particularly useful in my rapid prototyping projects! How about you?
karpathy 6 months ago prev next
I recently gave a talk on the benefits of automated ML model training and deployment, showcasing Kubeflow and other ML platforms. Video is available here:
ukasz 6 months ago prev next
Great presentation Karpathy! Glad you could include Kubeflow in the mix. I find their approach to ML development really refreshing!
nisha 6 months ago prev next
I think Kubeflow's user interface needs improvement. Have others experienced difficulty with the interface lately? I remember it being more intuitive before.
ukasz 6 months ago next
We use Kubeflow for our projects and find its interface quite comfortable to work with. There have been occasional bugs, but most have been resolved quickly.
5uperfa5t 6 months ago prev next
I'm looking forward to a curated list of use cases and templates for deployments with Kubeflow. Is there a repository for this kind of information yet?
igenious 6 months ago prev next
You may find this tutorial on deploying ML models using Kubeflow helpful: <https://tutorials.kubeflow.org/deployments/>
shilen 6 months ago prev next
Can the community create a catalog of official and unofficial, curated Kubeflow plugins? I'd be interested in seeing what others have come up with.
datawhisper 6 months ago prev next
Something I love about Kubeflow is the ability to smoothly migrate between different cloud platforms. Has anyone encountered any limitations?
mindw0rk 6 months ago prev next
If you liked the Karpathy presentation, you will also enjoy a discussion between him and other industry professionals on centralized ML platforms like Kubeflow.
mindw0rk 6 months ago next
@pragma This is the link to the roundtable I mentioned earlier, it includes Karpathy and other industry professionals: <https://youtu.be/Gg45FubgoOk>
pragma 6 months ago prev next
Kubeflow is a powerful solution for ML development, but it might be difficult for beginners to learn. Any resources you would recommend for getting started?
onejason 6 months ago prev next
I agree that Kubeflow can be intimidating to beginners. I've used this course to learn and suggest it as a great starting point for those interested: <https://www.udacity.com/course/deploying-machine-learning-models-with-kubeflow--ud086>
randomcoder 6 months ago prev next
Kubeflow seems perfect for rapid prototyping and experimentation, while providing a seamless transition to production environments.
saravana 6 months ago prev next
What are the best practices for setting up custom metrics inside Kubeflow? Is there a good guide that you've seen somewhere?
superkrish 6 months ago prev next
There's a helpful section on metrics and monitoring in the Kubeflow documentation, which you can access here: <https://www.kubeflow.org/docs/other-guides/metrics/>
codeguy 6 months ago prev next
Is anyone using Kubeflow with on-prem hardware? Interested to know how well it works in a local environment.
smiledog 6 months ago prev next
I've used Kubeflow with an on-premise setup and found it to function well as long as the hardware resources were sufficient. The community is also very supportive!