[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
[Device Plugins](https://www.tensorflow.org/install/gpu_plugins#available_devices).

A smaller CPU-only TensorFlow package is also available:

```
 pip install tensorflow-cpu
```

    name = "is_cuda_enabled",
    actual = if_oss(
        "@local_config_cuda//:is_cuda_enabled",
        "@local_config_cuda//cuda:using_config_cuda",
    ),
)

# Config setting that is satisfied when CUDA device code should be compiled
# with clang. It does not imply that CUDA support has been enabled.
alias(
    name = "is_cuda_compiler_clang",
    actual = if_oss(
        "@local_config_cuda//:is_cuda_compiler_clang",

        as arg with default value `None` to choose the I/O device for loading
        models and weights.
    *   Update `tf.saved_model.SaveOptions` with
        [`experimental_io_device`](https://www.tensorflow.org/versions/r2.3/api_docs/python/tf/saved_model/SaveOptions?hl=en)
        as arg with default value `None` to choose the I/O device for saving
        models and weights.

2. **Disclose significant AI involvement.** If AI tools played a major role in producing your contribution, say so in the description. We are not interested in incidental use, such as asking a chatbot for advice or accepting an autocomplete suggestion—but rather in cases where a tool generated substantial portions of the work. Disclosure will not reduce the likelihood of your PR or issue being accepted; we ask because information about AI tool use helps us...

common:mkl --define=build_with_mkl=true --define=enable_mkl=true
common:mkl --define=tensorflow_mkldnn_contraction_kernel=0
common:mkl --define=build_with_openmp=true
common:mkl -c opt

# config to build OneDNN backend with a user specified threadpool.
common:mkl_threadpool --define=build_with_mkl=true --define=enable_mkl=true
common:mkl_threadpool --define=tensorflow_mkldnn_contraction_kernel=0

                """
            ),
        ] = None,
        terms_of_service: Annotated[
            str | None,
            Doc(
                """
                A URL to the Terms of Service for your API.

                It will be added to the generated OpenAPI (e.g. visible at `/docs`).

                Read more at the

 * to be the last step of the pipeline.
 *
 * <h4>Transforming</h4>
 *
 * To derive the next step by asynchronously applying a function to an input step's value, call
 * {@link #transform(ClosingFunction, Executor)} or {@link #transformAsync(AsyncClosingFunction,
 * Executor)} on the input step.
 *
 * <h4>Catching</h4>
 *
 * To derive the next step from a failed input step, call {@link #catching(Class, ClosingFunction,

            "graph", "graphMain",
            "projectGraphReport",
            "ripples",
            "aggregateAdvice",
        ) -> true

        task.name.startsWithAnyOf(
            "advice",
            "analyzeClassUsage",
            "analyzeJar",
            "constantUsageDetector",
            "createVariantFiles",
            "findUnusedProcs",
            "generalsUsageDetector",

* **No serialization**: Just pass the raw, unstructured data around. Serialization is for people who don't trust their LLMs.
* **Embrace modern AI coding practices**: Leave everything up to an LLM to decide. The model knows best. Always.

        self.scheme_name = scheme_name or self.__class__.__name__
        self.auto_error = auto_error

    def make_not_authenticated_error(self) -> HTTPException:
        """
        The OAuth 2 specification doesn't define the challenge that should be used,
        because a `Bearer` token is not really the only option to authenticate.

        But declaring any other authentication challenge would be application-specific