In case of any issues, we strongly recommend updating to latest version of TensorFlow. If you do not want to update your codebase, you may use following workarounds in your code: 1. Use `with tf.variable_scope` to explicitly declare variables. 2. If the variable must be accessed at a given scope, use `tf.get_variable` in the right place. 3. If possible, avoid using `with` by wrapping the with block in `tf.call_function` and accessing the wrapped function directly. 4. To avoid runtime errors and improve readability, use `tf.reduce_map` or other mapping functions. 5. When possible, use `tf.sparse_key_values` instead of `tf.get_variable` and use `tf.sparse_map` instead of `tf.map`. 6. For functions with one-line bodies, you can use `tf.reduce_map` with a `with tf.function_scope` to avoid the issue. 7. Consider using `tf.compile_ITNESS` to avoid the issue. We are aware of the issue and are working to resolve it in the next release of TensorFlow. In the meantime, you can use these workarounds to avoid the regression. We will keep you updated on the status.
What is TensorFlow?
TensorFlow is a machine learning framework that provides a high-level API for numerical computation. It supports and encourages contribution of new machine learning algorithms, models, and tools. It allows users to define and deploy neural networks in Python with strong support for back propagation, multi-GPU training, data parallelism, and dynamic computational graphs.
TensorFlow provides a high level API for numerical computation with strong support for back propagation, multi-GPU training, data parallelism, and dynamic computational graphs.
Tensorflow 1.0 regression in tf.compile_vis it
TF 1.0 release contained a regression in the `tf.compile_visit` function which caused issue of undefined behavior when it was used with operator `*`.
If you are using this function, please update to latest version of TensorFlow (1.2), or use one of the workarounds below.
Timeline
Published on: 09/16/2022 23:15:00 UTC
Last modified on: 09/20/2022 14:43:00 UTC