Restrictions
- This is a specific guide for TensorFlow 1.15.
- Data types float64, complex64, complex128, and DT_VARIANT are not supported.
- Supported data formats include NCHW, NHWC, NC, HWCN, and CN.
- For condition branches and iteration branches, only tf.cond, tf.while_loop, and tf.case are supported.
- During multi-device training, NPURunConfig does not support save_checkpoints_secs in tf.estimator.RunConfig.
- During multi-device training, saving the summary information (via the tf.summary API) of only a single device is not supported.
- For the
Atlas training products , the operators do not support the Inf or NaN inputs. - During data preprocessing, only dataset and placeholder modes are supported for data reading; queue-based data reading is not supported.
- If you spawn processes using the Python package multiprocessing, you are advised to use the forkserver method as opposed to the fork method.
In Python versions 3.8 to 3.11, using the fork method may copy the lock state of the main process when a child process is created. If this child process subsequently attempts to acquire the lock, a deadlock may occur, which in turn causes the service process to hang.