Accelerator profiles and Habana integration - phase one - draft #45
Conversation
| * SciPy 1.10 | ||
| * Kfp-tekton 1.5 | ||
| * PyTorch 2.0 | ||
| * Elyra 3.15 |
There was a problem hiding this comment.
Would be also nice to have all these sorted alphabetically. But that's an optional thing and probably should be done in a separate commit. I'll try to propose this in upstream of the project too.
There was a problem hiding this comment.
Sure Jan, because of time constraints, we can do this in a forthcoming release. I'd prefer to create a separate Jira for this work, just to keep it separate from this.
modules/adding-a-model-server-to-your-data-science-project.adoc
Outdated
Show resolved
Hide resolved
chtyler
force-pushed
the
DS-11819-accelerator-profiles-phase-1
branch
from
2c71b07
to
8e84015
Compare
modules/adding-a-model-server-to-your-data-science-project.adoc
Outdated
Show resolved
Hide resolved
modules/adding-a-model-server-to-your-data-science-project.adoc
Outdated
Show resolved
Hide resolved
|
@FedeAlonso - please see the latest changes from the most recent commit. I have addressed your feedback. If you notice anything else, please let me know. @jstourac - your suggested changes should now be applied. |
|
|
||
| For accelerators that are new to your deployment, you must manually configure an accelerator profile for the accelerator. If your deployment contained NVIDIA GPUs before upgrading {productname-short}, an accelerator profile generates automatically after you upgrade to the latest version of {productname-short}, and resides in the `Instances` section of the `AcceleratorProfile` custom resource definition (CRD). | ||
|
|
||
| If you add NVIDIA GPUs to your deployment after upgrading {productname-short}, or you add them to your deployment after you install a new version of the {productname-short} Operator, you must manually create an accelerator profile for your NVIDIA GPUs. |
There was a problem hiding this comment.
1/ I think we should not mention NVIDIA GPU specifically in this doc unless we are giving some particular example. We should rather use some generic term like accelerator, hardware accelerator, hardware accelerator unit or similar?
2/ These two upgrade information - it's just a one time thing for the upgrade between two particular releases. We should put this note into some Upgrade Notes - that AcceleratorProfile will be created automatically if your deployment has had the hardware accelerator unit pre-configured before the upgrade already.
In general, here we should only have some message like this:
For hardware accelerators that are new to your deployment, you must manually configure an accelerator profile for that particular hardware accelerator. If your deployment contained a hardware accelerator before the upgrade, it will be preserved after the upgrade.
WDYT?
There was a problem hiding this comment.
I have reworded the parapgraph based on your suggestion. I have used the term "accelerators".
There was a problem hiding this comment.
I will update the downstream RHODS upgrade guides with information such as:
For hardware accelerators that are new to your deployment, you must manually configure an accelerator profile for that particular hardware accelerator. If your deployment contained a hardware accelerator before the upgrade, it will be preserved after the upgrade.
chtyler
force-pushed
the
DS-11819-accelerator-profiles-phase-1
branch
from
01bc844
to
ebbd80f
Compare
…accelerator profiles
…na operator - addressed further qe feedback
chtyler
force-pushed
the
DS-11819-accelerator-profiles-phase-1
branch
from
2c8386b
to
403a740
Compare
|
Merging. |
Description
Initial draft on accelerator profiles and Habana integration with ODH.
How Has This Been Tested?
Merge criteria: