Type DeviceAssignment
Namespace tensorflow.tpu.experimental
Parent PythonObjectContainer
Interfaces IDeviceAssignment
Mapping from logical cores in a computation to the physical TPU topology. Prefer to use the `DeviceAssignment.build()` helper to construct a
`DeviceAssignment`; it is easier if less flexible than constructing a
`DeviceAssignment` directly.
Methods
- build
- build_dyn
- coordinates
- coordinates_dyn
- host_device
- host_device_dyn
- lookup_replicas_dyn
- tpu_device
- tpu_device_dyn
- tpu_ordinal
- tpu_ordinal_dyn
Properties
Public instance methods
ValueTuple coordinates(int replica, int logical_core)
Returns the physical topology coordinates of a logical core.
object coordinates_dyn(object replica, object logical_core)
Returns the physical topology coordinates of a logical core.
string host_device(int replica, int logical_core, string job)
Returns the CPU device attached to a logical core.
object host_device_dyn(ImplicitContainer<T> replica, ImplicitContainer<T> logical_core, object job)
Returns the CPU device attached to a logical core.
object lookup_replicas_dyn(object task_id, object logical_core)
Lookup replica ids by task number and logical core.
Parameters
-
objecttask_id - TensorFlow task number.
-
objectlogical_core - An integer, identifying a logical core.
Returns
-
object - A sorted list of the replicas that are attached to that task and logical_core.
string tpu_device(int replica, int logical_core, object job)
Returns the name of the TPU device assigned to a logical core.
object tpu_device_dyn(ImplicitContainer<T> replica, ImplicitContainer<T> logical_core, object job)
Returns the name of the TPU device assigned to a logical core.
object tpu_ordinal(int replica, int logical_core)
Returns the ordinal of the TPU device assigned to a logical core.
object tpu_ordinal_dyn(ImplicitContainer<T> replica, ImplicitContainer<T> logical_core)
Returns the ordinal of the TPU device assigned to a logical core.
Public static methods
DeviceAssignment build(object topology, object computation_shape, object computation_stride, int num_replicas)
object build_dyn(object topology, object computation_shape, object computation_stride, ImplicitContainer<T> num_replicas)
Public properties
object core_assignment get;
The logical to physical core mapping.
object core_assignment_dyn get;
The logical to physical core mapping.
object num_cores_per_replica get;
The number of cores per replica.
object num_cores_per_replica_dyn get;
The number of cores per replica.
object num_replicas get;
The number of replicas of the computation.
object num_replicas_dyn get;
The number of replicas of the computation.
object PythonObject get;
object topology get;
A `Topology` that describes the TPU topology.
object topology_dyn get;
A `Topology` that describes the TPU topology.