Apply to speak at TensorFlow World. Deadline April 23rd. Propose talk



Defined in tensorflow/python/compat/

Return true if the forward compatibility window has expired.

See Version compatibility.

Forward-compatibility refers to scenarios where the producer of a TensorFlow model (a GraphDef or SavedModel) is compiled against a version of the TensorFlow library newer than what the consumer was compiled against. The "producer" is typically a Python program that constructs and trains a model while the "consumer" is typically another program that loads and serves the model.

TensorFlow has been supporting a 3 week forward-compatibility window for programs compiled from source at HEAD.

For example, consider the case where a new operation MyNewAwesomeAdd is created with the intent of replacing the implementation of an existing Python wrapper - tf.add. The Python wrapper implementation should change from something like:

def add(inputs, name=None):
  return gen_math_ops.add(inputs, name)


from tensorflow.python.compat import compat

def add(inputs, name=None):
  if compat.forward_compatible(year, month, day):
    # Can use the awesome new implementation.
    return gen_math_ops.my_new_awesome_add(inputs, name)
  # To maintain forward compatibiltiy, use the old implementation.
  return gen_math_ops.add(inputs, name)

Where year, month, and day specify the date beyond which binaries that consume a model are expected to have been updated to include the new operations. This date is typically at least 3 weeks beyond the date the code that adds the new operation is committed.


  • year: A year (e.g., 2018).
  • month: A month (1 <= month <= 12) in year.
  • day: A day (1 <= day <= 31, or 30, or 29, or 28) in month.


True if the caller can expect that serialized TensorFlow graphs produced can be consumed by programs that are compiled with the TensorFlow library source code after (year, month, day).