|---| | [ View source on GitHub](https://github.com/tensorflow/tensorflow/blob/v2.21.0/tensorflow/python/data/ops/debug_mode.py#L24-L72) | Enables debug mode for tf.data. #### View aliases **Compat aliases for migration** See [Migration guide](https://www.tensorflow.org/guide/migrate) for more details. [`tf.compat.v1.data.experimental.enable_debug_mode`](https://www.tensorflow.org/api_docs/python/tf/data/experimental/enable_debug_mode)
tf.data.experimental.enable_debug_mode() Example usage with pdb module: import tensorflow as tf import pdb tf.data.experimental.enable_debug_mode() def func(x): # Python 3.7 and older requires `pdb.Pdb(nosigint=True).set_trace()` pdb.set_trace() x = x + 1 return x dataset = tf.data.Dataset.from_tensor_slices([1, 2, 3]) dataset = dataset.map(func) for item in dataset: print(item) The effect of debug mode is two-fold: 1) Any transformations that would introduce asynchrony, parallelism, or non-determinism to the input pipeline execution will be forced to execute synchronously, sequentially, and deterministically. 2) Any user-defined functions passed into tf.data transformations such as `map` will be wrapped in [`tf.py_function`](https://www.tensorflow.org/api_docs/python/tf/py_function) so that their body is executed "eagerly" as a Python function as opposed to a traced TensorFlow graph, which is the default behavior. Note that even when debug mode is enabled, the user-defined function is still traced to infer the shape and type of its outputs; as a consequence, any `print` statements or breakpoints will be triggered once during the tracing before the actual execution of the input pipeline. > [!NOTE] > **Note:** As the debug mode setting affects the construction of the tf.data input pipeline, it should be enabled before any tf.data definitions.


| ## Raises || |---|---| | `ValueError` | When invoked from graph mode. |