# All required CUDA packages
cuda-compat-13-0
cuda-command-line-tools-13-0
cuda-cudart-dev-13-0
cuda-nvcc-13-0
cuda-cupti-13-0
cuda-nvprune-13-0
cuda-libraries-13-0
cuda-libraries-dev-13-0
cuda-nvml-dev-13-0
libcufft-13-0
libcurand-13-0
libcusolver-dev-13-0
libcusparse-dev-13-0
libcublas-13-0
libcublas-dev-13-0
libnccl-dev=2.27.7-1+cuda13.0
libnccl2=2.27.7-1+cuda13.0

# packages only provide the "0.10-2build1" version. We use patchelf to manipulate
# certain shared libraries during the wheel building process (https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tools/pip_package/build_pip_package.sh#L255-L262).
# When we use Patchelf versions <0.12, those shared libraries end up with a

# Tensorflow to the default, however test coverage wasn't enough to catch the
# errors.
# There is ongoing work on Bazel team's side to provide support for transitive
# shared libraries. As part of migrating to transitive shared libraries, we
# hope to provide a better mechanism for control over symbol exporting, and
# then tackle this issue again.
#
# TODO: Remove the following two lines once TF doesn't depend on Bazel wrapping

[TensorFlow](https://www.tensorflow.org/) is an end-to-end open source platform
for machine learning. It has a comprehensive, flexible ecosystem of
[tools](https://www.tensorflow.org/resources/tools),
[libraries](https://www.tensorflow.org/resources/libraries-extensions), and
[community](https://www.tensorflow.org/community) resources that lets
researchers push the state-of-the-art in ML and developers easily build and
deploy ML-powered applications.

# (e.g. _pywrap_tensorflow_internal.so for Python, libtensorflow.so for the C
# API) is responsible for registering ops with libtensorflow_framework.so. In
# addition to this core set of ops, user libraries which are loaded (via
# TF_LoadLibrary/tf.load_op_library) register their ops and kernels with this
# shared object directly.
#
# For example, from Python tf.load_op_library loads a custom op library (via

* Hermetic CUDA support is added.

  Hermetic CUDA uses a specific downloadable version of CUDA instead of the user’s locally installed CUDA. Bazel will download CUDA, CUDNN and NCCL distributions, and then use CUDA libraries and tools as dependencies in various Bazel targets. This enables more reproducible builds for Google ML projects and supported CUDA versions.