tensorflow/models · error · ValueError

When {num_gpus} GPUs are specified, distribution_strategy fl

Error message

When {num_gpus} GPUs are specified, distribution_strategy flag cannot be set to `off`.

What it means

Error "When {num_gpus} GPUs are specified, distribution_strategy flag cannot be set to `off`." thrown in tensorflow/models.

Source

Thrown at official/common/distribute_utils.py:144

      `distribution_strategy` is `tpu` but `tpu_address` is not specified.
  """
  del kwargs
  if num_gpus < 0:
    raise ValueError("`num_gpus` can not be negative.")

  if not isinstance(distribution_strategy, str):
    msg = ("distribution_strategy must be a string but got: %s." %
           (distribution_strategy,))
    if distribution_strategy == False:  # pylint: disable=singleton-comparison,g-explicit-bool-comparison
      msg += (" If you meant to pass the string 'off', make sure you add "
              "quotes around 'off' so that yaml interprets it as a string "
              "instead of a bool.")
    raise ValueError(msg)

  distribution_strategy = distribution_strategy.lower()
  if distribution_strategy == "off":
    if num_gpus > 1:
      raise ValueError(f"When {num_gpus} GPUs are specified, "
                       "distribution_strategy flag cannot be set to `off`.")
    # Return the default distribution strategy.
    return tf.distribute.get_strategy()

  if distribution_strategy == "tpu":
    # When tpu_address is an empty string, we communicate with local TPUs.
    # Bug workaround that in v5p we need to explicitly specify the device
    # assignment when using tpu strategy, adding device assignment to the
    # strategy.
    cluster_resolver = tf.distribute.cluster_resolver.TPUClusterResolver(
        tpu=tpu_address
    )
    if tpu_address not in ("", "local"):
      tf.config.experimental_connect_to_cluster(cluster_resolver)
    topology = tf.tpu.experimental.initialize_tpu_system(cluster_resolver)

    device_assignment = None
    if hasattr(tf.tpu.experimental, "HardWareFeature"):

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/common/distribute_utils.py:144 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/57126fbdf06d0dd4. Report an issue: GitHub.