TensorFlow 2.0 Beta is available Learn more

tf.contrib.tpu.while_loop

View source on GitHub

Builds a training loop for TPUs.

tf.contrib.tpu.while_loop(
    condition,
    body,
    inputs=None,
    infeed_queue=None,
    name=None
)

The set of loop-carried tensors corresponds to inputs. Both condition and body take the current value of the loop-carried tensors. 'body' additionally takes a tuple of infeed from infeed_queue if infeed_queue is not None. condition must return a single boolean value that determines whether iteration continues. body must return an updated list of values for the loop-carried tensors.

Args:

  • condition: a Python function that builds the loop condition.
  • body: a Python function that builds the loop body.
  • inputs: a list of initial values passed into the training loop, or None (equivalent to an empty list).
  • infeed_queue: if not None, the infeed queue from which to append a tuple of arguments as inputs to condition.
  • name: (Deprecated) Does nothing.

Returns:

The final values of the loop-carried tensors.

Raises:

  • TypeError: if body or condition has the wrong signature.