# bazel test invocation as normal.
test:nonpip_filters --test_tag_filters=-no_oss,-oss_serial,-gpu,-tpu,-benchmark-test,-v1only,-no_aarch64,-no_oss_py38,-no_oss_py39,-no_oss_py310
test:nonpip_filters --build_tag_filters=-no_oss,-oss_serial,-gpu,-tpu,-benchmark-test,-v1only,-no_aarch64,-no_oss_py38,-no_oss_py39,-no_oss_py310

# bazel test invocation as normal.
test:nonpip_filters --test_tag_filters=-no_oss,-oss_serial,-gpu,-tpu,-benchmark-test,-v1only,-no_aarch64,-no_oss_py38,-no_oss_py39,-no_oss_py310
test:nonpip_filters --build_tag_filters=-no_oss,-oss_serial,-gpu,-tpu,-benchmark-test,-v1only,-no_aarch64,-no_oss_py38,-no_oss_py39,-no_oss_py310

source ci/official/envs/linux_x86_build
TFCI_BAZEL_COMMON_ARGS="--repo_env=HERMETIC_PYTHON_VERSION=$TFCI_PYTHON_VERSION --config release_cpu_linux --config=tpu"
TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_tpu
TFCI_BUILD_PIP_PACKAGE_ARGS="--repo_env=WHEEL_NAME=tensorflow_tpu"
TFCI_LIB_SUFFIX="-tpu-linux-x86_64"
TFCI_WHL_BAZEL_TEST_ENABLE=0
TFCI_WHL_IMPORT_TEST_ENABLE=0
TFCI_WHL_SIZE_LIMIT=580M

#     dbg:              Build with debug info
#
# TF version options;
#     v2: Build TF v2
#
# Feature and Third party library support options:
#     xla:          Build TF with XLA
#     tpu:          Build TF with TPU support
#     cuda:         Build with CUDA support.
#     cuda_clang    Build with CUDA Clang support.
#     rocm:         Build with AMD GPU support (rocm)
#     mkl:          Enable full mkl support.

source ci/official/envs/linux_x86
TFCI_BAZEL_COMMON_ARGS="--repo_env=HERMETIC_PYTHON_VERSION=$TFCI_PYTHON_VERSION --config release_cpu_linux --config=tpu"
TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_tpu
TFCI_BUILD_PIP_PACKAGE_ARGS="--repo_env=WHEEL_NAME=tensorflow_tpu"
TFCI_LIB_SUFFIX="-tpu-linux-x86_64"
TFCI_WHL_BAZEL_TEST_ENABLE=0
TFCI_WHL_IMPORT_TEST_ENABLE=0
TFCI_WHL_SIZE_LIMIT=580M

         * It has an added advantage since you can you easily switch to different hardware accelerators (cpu, gpu, tpu) as per the task.
         * All you need is a good internet connection and you are all set.
      * Try to build TF from sources by changing CPU optimization flags.
   
   *Please let us know if this helps.*
   

      TFE_NewContext(opts.get(), status.get()), TFE_DeleteContext);
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  // Skip the test if no TPU is available.
  std::unique_ptr<TF_DeviceList, decltype(&TF_DeleteDeviceList)> devices(
      TFE_ContextListDevices(context.get(), status.get()), TF_DeleteDeviceList);

config_setting(
    name = "disable_mlir_bridge",
    define_values = {"enable_mlir_bridge": "false"},
    visibility = ["//visibility:public"],
)

# This flag enables experimental TPU support
bool_flag(
    name = "enable_tpu_support",
    build_setting_default = False,
)

config_setting(
    name = "with_tpu_support_define",
    define_values = {"with_tpu_support": "true"},

*   `tf.tpu.experimental.embedding`:

    *   `tf.tpu.experimental.embedding.FeatureConfig` now takes an additional
        argument `output_shape` which can specify the shape of the output
        activation for the feature.
    *   `tf.tpu.experimental.embedding.TPUEmbedding` now has the same behavior
        as `tf.tpu.experimental.embedding.serving_embedding_lookup` which can

    const TFE_OpAttrs* attributes, int expected_max_outputs,
    TF_Status* status) {
  absl::optional<std::vector<MaybeParallelTensorOwned>> result;
  // TODO(allenl): We should remove "TPU" from these op names at the very least,
  // or consider other ways of packing/unpacking parallel tensors.
  if (operation_name == std::string("TPUReplicatedInput")) {