[ci/docker/air] Update ML/DL dependencies to latest releases / Remove Py3.6 Docker images#28808
Conversation
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
…ir-update-dependencies
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
…endencies # Conflicts: # python/requirements/ml/requirements_rllib.txt
Signed-off-by: Kai Fricke <kai@anyscale.com>
|
Tests are passing except for some in RLlib - I will work with the rllib team to resolve this. But the rest is ready for review cc @pcmoritz @richardliaw for general testing infrastructure The base CI GPU image has been updated to support CUDA 11.6 already. |
|
I guess this PR could incorporate, supercede and close #26072? |
# Conflicts: # docker/ray-ml/Dockerfile # python/requirements/ml/requirements_tune.txt
@ddelange how about we do this in two steps - merge this first and land your changes once pyg-team/pyg-lib#66 is addressed? |
|
totally fine by me :) |
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Artur Niederfahrenhorst <artur@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
Signed-off-by: Kai Fricke <kai@anyscale.com>
# Conflicts: # rllib/tests/run_regression_tests.py
|
|
||
| # Test against all frameworks. | ||
| for _ in framework_iterator(config, with_eager_tracing=True): | ||
| for _ in framework_iterator(config, with_eager_tracing=False): |
There was a problem hiding this comment.
@gjoliver @kouroshHakha Not sure why this test did not fail initially, but A3C seems error with eager tracing as well, right @krfricke ?
… Py3.6 Docker images (ray-project#28808) This PR upgrades some of our dependencies to the latest releases. At the moment, we are testing with a lower bound of dependencies. This is good as we ensure compatibility with these versions, but at the same time we don't test compatibility with more recent releases. We prioritize backwards compatibility over compatibility with more recent versions. This PR: - Introduces a set of _legacy dependencies_ (the current versions in the CI). - This is a lower bound of dependencies that we guarantee compatibility with - The regular dependencies are upgraded to more recent releases - We remove Docker builds for Python 3.6. as this is incompatible with more recent versions of tensorflow/torch - Wheel building for 3.6 is not affected. Signed-off-by: Kai Fricke <kai@anyscale.com> Signed-off-by: Artur Niederfahrenhorst <artur@anyscale.com> Co-authored-by: Artur Niederfahrenhorst <artur@anyscale.com> Signed-off-by: Weichen Xu <weichen.xu@databricks.com>
Why are these changes needed?
This PR upgrades some of our dependencies to the latest releases.
At the moment, we are testing with a lower bound of dependencies. This is good as we ensure compatibility with these versions, but at the same time we don't test compatibility with more recent releases. We prioritize backwards compatibility over compatibility with more recent versions.
This PR:
Related issue number
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.