#   -r ci/official/requirements_updater/nvidia-constraints.txt
    #   -r ci/official/requirements_updater/nvidia-requirements.txt
    #   nvidia-cudnn-cu12
    #   nvidia-cusolver-cu12
nvidia-cuda-cupti-cu12==12.5.82 \
    --hash=sha256:4f835281cf492e2bedd153f5c3de9da8f1d775a419468305e64ce73b3b0c6dc3 \
    --hash=sha256:bde77a5feb66752ec61db2adfe47f56b941842825b4c7e2068aff27c9d107953 \

    #   -r ci/official/requirements_updater/nvidia-constraints.txt
    #   -r ci/official/requirements_updater/nvidia-requirements.txt
    #   nvidia-cudnn-cu12
    #   nvidia-cusolver-cu12
nvidia-cuda-cupti-cu12==12.5.82 \
    --hash=sha256:4f835281cf492e2bedd153f5c3de9da8f1d775a419468305e64ce73b3b0c6dc3 \
    --hash=sha256:bde77a5feb66752ec61db2adfe47f56b941842825b4c7e2068aff27c9d107953 \

    #   -r ci/official/requirements_updater/nvidia-constraints.txt
    #   -r ci/official/requirements_updater/nvidia-requirements.txt
    #   nvidia-cudnn-cu12
    #   nvidia-cusolver-cu12
nvidia-cuda-cupti-cu12==12.5.82 \
    --hash=sha256:4f835281cf492e2bedd153f5c3de9da8f1d775a419468305e64ce73b3b0c6dc3 \
    --hash=sha256:bde77a5feb66752ec61db2adfe47f56b941842825b4c7e2068aff27c9d107953 \

* 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.

[Docker container](https://www.tensorflow.org/install/docker), and
[build from source](https://www.tensorflow.org/install/source).

To install the current release, which includes support for
[CUDA-enabled GPU cards](https://www.tensorflow.org/install/gpu) *(Ubuntu and
Windows)*:

```
 pip install tensorflow
```

Other devices (DirectX and MacOS-metal) are supported using

# 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

# All required CUDA packages
cuda-compat-12-1
cuda-command-line-tools-12-1
cuda-cudart-dev-12-1
cuda-nvcc-12-1
cuda-cupti-12-1
cuda-nvprune-12-1
cuda-libraries-12-1
cuda-libraries-dev-12-1
cuda-nvml-dev-12-1
libcufft-12-1
libcurand-12-1
libcusolver-dev-12-1
libcusparse-dev-12-1
libcublas-12-1
libcublas-dev-12-1
libnccl-dev=2.18.3-1+cuda12.1
libnccl2=2.18.3-1+cuda12.1

# All required CUDA packages
cuda-compat-12-3
cuda-command-line-tools-12-3
cuda-cudart-dev-12-3
cuda-nvcc-12-3
cuda-cupti-12-3
cuda-nvprune-12-3
cuda-libraries-12-3
cuda-libraries-dev-12-3
cuda-nvml-dev-12-3
libcufft-12-3
libcurand-12-3
libcusolver-dev-12-3
libcusparse-dev-12-3
libcublas-12-3
libcublas-dev-12-3
libnccl-dev=2.19.3-1+cuda12.3
libnccl2=2.19.3-1+cuda12.3

# All required CUDA packages
cuda-compat-12-1
cuda-command-line-tools-12-1
cuda-cudart-dev-12-1
cuda-nvcc-12-1
cuda-cupti-12-1
cuda-nvprune-12-1
cuda-libraries-12-1
cuda-libraries-dev-12-1
cuda-nvml-dev-12-1
libcufft-12-1
libcurand-12-1
libcusolver-dev-12-1
libcusparse-dev-12-1
libcublas-12-1
libcublas-dev-12-1
libnccl-dev=2.18.3-1+cuda12.1
libnccl2=2.18.3-1+cuda12.1

3. Copie o arquivo com os requisitos para o diretório `/code`.

    Copie **somente** o arquivo com os requisitos primeiro, não o resto do código.

    Como esse arquivo **não muda com frequência**, o Docker irá detectá-lo e usar o **cache** para esse passo, habilitando o cache para o próximo passo também.

4. Instale as dependências de pacote vindas do arquivo de requisitos.