[pytorch] | [build, test] | [ec2, ecs, eks, sagemaker] PT 1.4 patch release #216
Conversation
| account_id: &ACCOUNT_ID <set-$ACCOUNT_ID-in-environment> | ||
| region: ®ION <set-$REGION-in-environment> | ||
| framework: &FRAMEWORK pytorch | ||
| version: &VERSION 1.5.0 |
There was a problem hiding this comment.
Our main buildspec file will be for 1.5 only.
Changes in 1.4 will far and fewer than PT1.5.
Please create a sperate buildspce for PT1.4 not PT1.5
There was a problem hiding this comment.
What I want to do is to use it as a one time release purpose file. Will recover to PT1.5 once PT 1.4 patch release done.
Which method do you prefer? To do modification on root buildspec.yml? It also requires a recover operation
There was a problem hiding this comment.
For now, this is fine. Let's use buildspec.yml for containers we want to build currently.
There was a problem hiding this comment.
Thanks for confirming.
|
|
||
| # Add any script or repo as required | ||
| RUN cd /var && wget https://raw.githubusercontent.com/kubeflow/pytorch-operator/master/examples/mnist/mnist.py | ||
| COPY mnist.py /var/mnist.py |
There was a problem hiding this comment.
This is introduced since PT 1.5. PT team change the script and we now maintain it as a separate artifact https://github.com/aws/deep-learning-containers/blob/master/pytorch/training/docker/build_artifacts/mnist.py
There was a problem hiding this comment.
Is this necessary now that we are building a PT 1.4 image?
There was a problem hiding this comment.
Talked with Tushar. We think it should be a general change and should also apply to PT1.4. @TusharKanekiDey
| LOG_DIR=${HOME_DIR}/logs | ||
|
|
||
| git clone https://github.com/pytorch/examples.git ${HOME_DIR}/artifacts/examples | ||
| sed -i "s/param.add_(-0.1 \* param.grad)/param.data.add_(-0.1 \* param.grad)/g" ${HOME_DIR}/artifacts/examples/regression/main.py # temporary patch for isssue https://github.com/pytorch/examples/issues/782 |
There was a problem hiding this comment.
There is a issue with this test. The PT example repo fixed a similar issue, but seems forgot to fix this file. pytorch/examples#779
I opened a PR for their repo but did not have response yet. This is a temp approach to make the test pass.
4 participants
Issue #, if available:
Checklist
Description:
Tests run:
DLC image/dockerfile:
Additional context:
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license. I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.