View source on GitHub |
Shuts down the TPU devices.
tf.tpu.experimental.shutdown_tpu_system(
cluster_resolver=None
)
This will clear all caches, even those that are maintained through sequential calls to tf.tpu.experimental.initialize_tpu_system, such as the compilation cache.
cluster_resolver
: A tf.distribute.cluster_resolver.TPUClusterResolver,
which provides information about the TPU cluster.RuntimeError
: If no TPU devices found for eager execution or if run in a
tf.function.