1:42 p.m.
Hello all,
First of all, I appreciate your work on this project and thank you for your
persistence.
I'm currently trying to use a mix of ODH/Kubeflow tools to cover a complete
AI infrastructure. I'm still a little bit confused about which components
to use :
- airflow vs kubeflow pipelines
- seldon core vs kubeflow kfserving
- kubeflow's jupyterlab vs odh's jupyterhub notebooks
- etc.
To help me make my decision, I'd like to know which are the components that
you guys think are stable enough.
For example, I think the airflow operator is still in v1alpha version and
the UI is not yet accessible from odh dashboard and thus not yet suited for
production, is it?
If you can guide me to where I can find useful information I'd appreciate
it.
Thanks in advance.
9:08 a.m.
Hi Tarik,
Excellent questions. From a stability perspective, all components are open
source and supported by their respective open source communities. In Open
Data Hub we do have multiple components offering similar functionalities in
the AI/ML phase and are driven to upgrades based on our user and community
interests. Airflow in our community is not as popular as Kubeflow pipelines
and probably needs an update, not being accessible from the dashboard is
not an indication of maturity, it should be there, please open an issue if
it is not showing up.
As for choosing which components to use, my advice to you is to start by
understanding the requirements of your users, your team and the cluster
available. Even though some of these components provide similar
functionality, digging deeper you will find different features. For example
while both Seldon and KFServing provide model serving, Seldon allows for
custom model metrics reporting for prometheus and grafana while KFServing
does not. On the other hand, KFServing uses Knative that includes features
such as AutoScaling, Canary rollouts and batching not present in Seldon.
I think providing documentation that illustrates the different features of
each component is a great idea.
Juana
On Wed, Nov 24, 2021 at 2:12 PM tarik lwa3er <itmwiw(a)gmail.com> wrote:
Hello all,
First of all, I appreciate your work on this project and thank you for
your persistence.
I'm currently trying to use a mix of ODH/Kubeflow tools to cover a
complete AI infrastructure. I'm still a little bit confused about
which components to use :
- airflow vs kubeflow pipelines
- seldon core vs kubeflow kfserving
- kubeflow's jupyterlab vs odh's jupyterhub notebooks
- etc.
To help me make my decision, I'd like to know which are the components
that you guys think are stable enough.
For example, I think the airflow operator is still in v1alpha version and
the UI is not yet accessible from odh dashboard and thus not yet suited for
production, is it?
If you can guide me to where I can find useful information I'd appreciate
it.
Thanks in advance.
_______________________________________________
Users mailing list -- users(a)lists.opendatahub.io
To unsubscribe send an email to users-leave(a)lists.opendatahub.io
5:42 p.m.
Ok I'll start by the requirements and start comparing the tools on a later
step. Thanks Juana for your reply.
Le lun. 29 nov. 2021 à 16:09, Juana Nakfour <jnakfour(a)redhat.com> a écrit :
Hi Tarik,
Excellent questions. From a stability perspective, all components are open
source and supported by their respective open source communities. In Open
Data Hub we do have multiple components offering similar functionalities in
the AI/ML phase and are driven to upgrades based on our user and community
interests. Airflow in our community is not as popular as Kubeflow pipelines
and probably needs an update, not being accessible from the dashboard is
not an indication of maturity, it should be there, please open an issue if
it is not showing up.
As for choosing which components to use, my advice to you is to start by
understanding the requirements of your users, your team and the cluster
available. Even though some of these components provide similar
functionality, digging deeper you will find different features. For example
while both Seldon and KFServing provide model serving, Seldon allows for
custom model metrics reporting for prometheus and grafana while KFServing
does not. On the other hand, KFServing uses Knative that includes features
such as AutoScaling, Canary rollouts and batching not present in Seldon.
I think providing documentation that illustrates the different features of
each component is a great idea.
Juana
On Wed, Nov 24, 2021 at 2:12 PM tarik lwa3er <itmwiw(a)gmail.com> wrote:
> Hello all,
> First of all, I appreciate your work on this project and thank you for
> your persistence.
> I'm currently trying to use a mix of ODH/Kubeflow tools to cover a
> complete AI infrastructure. I'm still a little bit confused about
> which components to use :
> - airflow vs kubeflow pipelines
> - seldon core vs kubeflow kfserving
> - kubeflow's jupyterlab vs odh's jupyterhub notebooks
> - etc.
> To help me make my decision, I'd like to know which are the components
> that you guys think are stable enough.
> For example, I think the airflow operator is still in v1alpha version and
> the UI is not yet accessible from odh dashboard and thus not yet suited for
> production, is it?
> If you can guide me to where I can find useful information I'd appreciate
> it.
> Thanks in advance.
> _______________________________________________
> Users mailing list -- users(a)lists.opendatahub.io
> To unsubscribe send an email to users-leave(a)lists.opendatahub.io
>
897
days inactive
903
days old
2 comments
2 participants
participants (2)
-
Juana Nakfour -
tarik lwa3er