tff.framework.local_cpp_executor_factory

Returns an execution context backed by C++ runtime.

tff.framework.local_cpp_executor_factory(
    *,
    default_num_clients: int = 0,
    max_concurrent_computation_calls: int = -1,
    stream_structs: bool = False,
    server_mesh: Optional[tf.experimental.dtensor.Mesh] = None,
    client_mesh: Optional[tf.experimental.dtensor.Mesh] = None
) -> tff.framework.ExecutorFactory

Args
`default_num_clients`	The number of clients to use as the default cardinality, if thus number cannot be inferred by the arguments of a computation.
`max_concurrent_computation_calls`	The maximum number of concurrent calls to a single computation in the C++ runtime. If nonpositive, there is no limit.
`stream_structs`	The flag to enable decomposing and streaming struct values.
`server_mesh`	If present, worker binary will create DTensor based executor on the server side using given serialized DTensor Mesh.
`client_mesh`	If present, worker binary will create DTensor based executor on the client side using given serialized DTensor Mesh.

Raises
`RuntimeError`	If an internal C++ worker binary can not be found.

tff.framework.local_cpp_executor_factory

Args

Raises