Upgrade PyTorch version to v2.2.1 #2279
Conversation
google-oss-prow
bot
requested review from
andreyvelich,
anencore94 and
johnugeorge
|
We need to restart the Anyway, this PR is ready for the review. PTAL /assign @andreyvelich @johnugeorge |
tenzen-y
changed the title
| torch==1.13.1 | ||
| torchvision==0.14.1 | ||
| torch==2.2.1 | ||
| torchvision==0.17.1 |
There was a problem hiding this comment.
Should we also update the PyTorch base image in our SDK: https://github.com/kubeflow/katib/blob/master/sdk/python/v1beta1/kubeflow/katib/constants/constants.py#L55 ?
| @@ -1,4 +1,4 @@ | |||
| cloudml-hypertune==0.1.0.dev6 | |||
| torch==1.13.1 | |||
| torchvision==0.14.1 | |||
| torch==2.2.1 | |||
There was a problem hiding this comment.
Do we need to install torch version in requirements if we already using nvcr.io/nvidia/pytorch:24.01-py3 image ?
There was a problem hiding this comment.
I guess that it is no need, but I'm not sure that having separate requirements.txt would be worth it. @andreyvelich WDYT?
There was a problem hiding this comment.
@tenzen-y We are still installing cloudml-hypertune and Pillow in the requirements file, so I think we can keep it.
There was a problem hiding this comment.
Ah, I see. You meant that we should install torch and torch vision via Dockerfile, instead of requirements.txt, right?
There was a problem hiding this comment.
I thought torch package is already installed as part of this image nvcr.io/nvidia/pytorch:24.01-py3, isn't ?
I am not sure about torchvision tho
There was a problem hiding this comment.
Yes, both torch and torchvisoin are packaged in NVIDIA image.
Let me create a separate requirements.txt.
There was a problem hiding this comment.
I think you misunderstood me.
Since torch and torchvision is installed in those image we can remove them from requirements.txt and install only cloudml-hypertune and Pillow. We don't need to separate requirements for CPU and GPU images.
Also, we can remove torch installation from examples/v1beta1/trial-images/darts-cnn-cifar10/requirements.txt
tenzen-y
force-pushed
the
upgrade-pytorch-version
branch
2 times, most recently
from
054c7a8
to
de4ef3f
Compare
| RUN pip install --prefer-binary --no-cache-dir torch==2.2.1 torchvision==0.17.1 | ||
| RUN pip install --prefer-binary --no-cache-dir -r requirements.txt |
There was a problem hiding this comment.
@andreyvelich Does this meet your intention?
Signed-off-by: Yuki Iwai <yuki.iwai.tz@gmail.com>
tenzen-y
force-pushed
the
upgrade-pytorch-version
branch
from
de4ef3f
to
e5c0e4b
Compare
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: andreyvelich, tenzen-y The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Thank you for the helping! |
|
@andreyvelich Could you restart the "Go Test / Unit Test (1.26.1) (pull_request) "? |
What this PR does / why we need it:
I upgraded PyTorch version to v2.2.1.
This is a blocker of #2278
Which issue(s) this PR fixes (optional, in
fixes #<issue number>(, fixes #<issue_number>, ...)format, will close the issue(s) when PR gets merged):Fixes #
Checklist: