Documentation

Restrictions

  1. This is a specific guide for TensorFlow 1.15. For TensorFlow 2.6.5 model porting instructions, see TensorFlow 2.6.5 Model Porting Guide.
  2. Data types float64, complex64, complex128, and DT_VARIANT are not supported.
  3. Supported data formats include NCHW, NHWC, NC, HWCN, and CN.
  4. For condition branches and iteration branches, only tf.cond, tf.while_loop, and tf.case are supported.
  5. During multi-device training, NPURunconfig does not support save_checkpoints_secs in tf.estimator.RunConfig.
  6. During multi-device training, it is not allowed to save the summary information (tf.summary) of only a single device.
  7. For the Atlas Training Series Product, the operators do not support the Inf or NaN inputs.
  8. For data preprocessing, data cannot be read in queue mode. Only dataset and placeholder modes are supported.
  9. If you spawn processes using the Python package multiprocessing, you are advised to use the forkserver method as opposed to the fork method.