Public API for tf._api.v2.experimental.dtensor namespace ## Classes [`class DTensorCheckpoint`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/DTensorCheckpoint): Manages saving/restoring trackable values to disk, for DTensor. (deprecated) [`class DTensorDataset`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/DTensorDataset): A dataset of DTensors. [`class DVariable`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/DVariable): A replacement for tf.Variable which follows initial value placement. [`class Layout`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/Layout): Represents the layout information of a DTensor. [`class Mesh`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/Mesh): Represents a Mesh configuration over a certain list of Mesh Dimensions. ## Functions [`barrier(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/barrier): Runs a barrier on the mesh. [`call_with_layout(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/call_with_layout): Calls a function in the DTensor device scope if `layout` is not None. [`check_layout(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/check_layout): Asserts that the layout of the DTensor is `layout`. [`client_id(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/client_id): Returns this client's ID. [`copy_to_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/copy_to_mesh): Copies a tf.Tensor onto the DTensor device with the given layout. [`create_distributed_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/create_distributed_mesh): Creates a distributed mesh. [`create_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/create_mesh): Creates a single-client mesh. [`create_tpu_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/create_tpu_mesh): Returns a distributed TPU mesh optimized for AllReduce ring reductions. [`default_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/default_mesh): Sets the default DTensor device mesh to use for enclosed functions. [`device_name(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/device_name): Returns the singleton DTensor device's name. [`enable_save_as_bf16(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/enable_save_as_bf16): Allows float32 DVariables to be checkpointed and restored as bfloat16. [`fetch_layout(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/fetch_layout): Fetches the layout of a DTensor. [`full_job_name(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/full_job_name): Returns the fully qualified TF job name for this or another task. [`get_default_mesh(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/get_default_mesh): Return the default mesh under the current dtensor device context. [`heartbeat_enabled(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/heartbeat_enabled): Returns true if DTensor heartbeat service is enabled. [`initialize_accelerator_system(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/initialize_accelerator_system): Initializes accelerators and communication fabrics for DTensor. [`initialize_multi_client(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/initialize_accelerator_system): Initializes accelerators and communication fabrics for DTensor. [`initialize_tpu_system(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/initialize_accelerator_system): Initializes accelerators and communication fabrics for DTensor. [`is_dtensor(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/is_dtensor): Check whether the input tensor is a DTensor. [`job_name(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/job_name): Returns the job name used by all clients in this DTensor cluster. [`jobs(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/jobs): Returns a list of job names of all clients in this DTensor cluster. [`local_devices(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/local_devices): Returns a list of device specs configured on this client. [`name_based_restore(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/name_based_restore): Restores from checkpoint_prefix to name based DTensors. [`name_based_save(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/name_based_save): Saves name based Tensor into a Checkpoint. [`num_clients(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/num_clients): Returns the number of clients in this DTensor cluster. [`num_global_devices(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/num_global_devices): Returns the number of devices of device_type in this DTensor cluster. [`num_local_devices(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/num_local_devices): Returns the number of devices of device_type configured on this client. [`pack(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/pack): Packs [`tf.Tensor`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/Tensor) components into a DTensor. [`preferred_device_type(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/preferred_device_type): Returns the preferred device type for the accelerators. [`relayout(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/relayout): Changes the layout of `tensor`. [`relayout_like(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/relayout_like): Changes the layout of `tensor` to the same as `layout_tensor`. [`run_on(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/run_on): Runs enclosed functions in the DTensor device scope. (deprecated) [`sharded_save(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/sharded_save): Saves given named tensor slices in a sharded, multi-client safe fashion. [`shutdown_accelerator_system(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/shutdown_accelerator_system): Shuts down the accelerator system. [`shutdown_tpu_system(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/shutdown_accelerator_system): Shuts down the accelerator system. [`unpack(...)`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/experimental/dtensor/unpack): Unpacks a DTensor into [`tf.Tensor`](https://www.tensorflow.org/versions/r2.15/api_docs/python/tf/Tensor) components.


| ## Other Members || |---|---| | MATCH | `'match'` | | UNSHARDED | `'unsharded'` |