-
Notifications
You must be signed in to change notification settings - Fork 2.2k
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
Fix build with latest TF nightly #3864
Conversation
Signed-off-by: Abolfazl Shahbazi <[email protected]>
Unit Test Results 1 193 files + 222 1 193 suites +222 13h 38m 53s ⏱️ + 2h 0m 0s For more details on these failures, see this check. Results for commit e9c76ec. ± Comparison against base commit 9b42eda. ♻️ This comment has been updated with latest results. |
Unit Test Results (with flaky tests) 1 313 files + 149 1 313 suites +149 15h 9m 49s ⏱️ + 2h 24m 25s For more details on these failures, see this check. Results for commit e9c76ec. ± Comparison against base commit 9b42eda. ♻️ This comment has been updated with latest results. |
Signed-off-by: Max H. Gerlach <[email protected]>
Thanks for the fix, @ashahba! I went ahead and added a commit to the PR branch so compatibility with TF <= 2.12 will be retained. |
Thanks @maxhgerlach for timely respond. |
After yesterday's PyTorch release we pick up
I can only find earlier Probably they stopped building for CUDA 11.6. There are newer packages |
On tf-nightly this allows to use an overloaded bool operator==(const ::tensorflow::error::Code& c1, const absl::StatusCode& c2) for backward compatibility. Signed-off-by: Max H. Gerlach <[email protected]>
Signed-off-by: Max H. Gerlach <[email protected]>
I've added another fix for tf-head (related to a refactoring towards |
Thanks @maxhgerlach but I'm seeing the I'm thinking that we should probably switch to |
I think the readthedocs deployment was broken for a while, several similar issues: readthedocs/readthedocs.org#10173 Edit: I don't see an easy way to trigger a new doc build from GitHub. We should just ignore the (transient) problem for this PR that doesn't change any docs. |
The build works fine now and tests pass. Thanks for kicking this off, @ashahba! Here are the (nightly) package version used by the pipelines:
Could you look over the C++ changes made in the PR, @romerojosh? |
Thanks @maxhgerlach |
All right, but I think it was really OK as it was. All tests passed. The docs build failure was just due to transient flakiness over at Readthedocs. |
Thanks for the approval and merge, @nvcastet! 👍 |
Thanks you all! |
Signed-off-by: Abolfazl Shahbazi [email protected]
This PR fixes
horovod
install issue for nightly TensorFlow as described in #3861Here are the steps to get a passing install:
Then apply the patch provided by this PR and run the installation as below:
python3 -m pip install --no-cache-dir -v -e .
I've only tested this on the following configuration: