Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

remove uses of deprecated torch API (removed on commit #33376 on pyto… #1731

Merged
merged 4 commits into from
Feb 25, 2020

Conversation

aavbsouza
Copy link
Contributor

The commit pytorch/pytorch#33376 removed the functions of the family THCState_getCurrentStream. The functionality of these functions are now provided by functions of the pytorch's C10 library.

These changes causes problems on the file read_events.cc that use some of these functions. I have replaced these functions with the C10 correspondents. However for backward compatibility the previously used functions still on the code guarded by a TORCH_VERSION check.

I have closed this same issue and I am now reopening it. Because accidentally I have messed up the signoff procedure when I first opened it. Sorry for the inconvenience

Signed-off-by: aavbsouza aavbsouza@gmail.com

@tgaddair
Copy link
Collaborator

Hey @aavbsouza, thanks for this fix! Can you rebase this change off of master? Now that #1733 has landed, we can test these changes on GPUs again.

Also, as part of your PR, can you revert the most recent change to this line to unpin the torch version, like so:

pip install --pre torch ${TORCHVISION_PACKAGE} -v -f https://download.pytorch.org/whl/nightly/cu${PYTORCH_CUDA}/torch_nightly.html; \

That way, we can verify that these changes fix compatibility issues with the latest PyTorch nightly build.

…rch/pytorch)

Signed-off-by: aavbsouza <aavbsouza@gmail.com>
Signed-off-by: aavbsouza <aavbsouza@gmail.com>
Signed-off-by: aavbsouza <aavbsouza@gmail.com>
@aavbsouza
Copy link
Contributor Author

Observing the results of the buildkite. First tree jobs fails, with all fails related to errors in getting the latest version of Tensorflow, for instance:

ERROR: Could not find a version that satisfies the requirement tf-nightly==2.1.0.dev20191203 (from versions: none)
ERROR: No matching distribution found for tf-nightly==2.1.0.dev20191203
ERROR: Service 'test-cpu-openmpi-py2_7-tfhead-kerashead-torchhead-mxnethead-pyspark2_4_0' failed to build: The command '/bin/bash -cu pip install ${TENSORFLOW_PACKAGE}' returned a non-zero code: 1

Some time after all jobs fails without producing any output

@tgaddair
Copy link
Collaborator

@aavbsouza it looks like TensorFlow may have purged some packages from pip. Instead of tf-nightly and tf-nightly-gpu, you can try setting tensorflow==2.1.0 and tensorflow-gpu==2.1.0 for now in docker-compose.test.yml.

Signed-off-by: aavbsouza <aavbsouza@gmail.com>
@tgaddair
Copy link
Collaborator

LGTM! @romerojosh can you review as well?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Development

Successfully merging this pull request may close these issues.

None yet

3 participants