3 ms·
tf.while code will run in parallel as far as possible (if it does not depend on each other) (up to some configurable degree).
by albertzeyer 8y ago
tf.while code will run in parallel as far as possible (if it does not depend on each other) (up to some configurable degree).
- duckworthd 8y agoYes! Though this answer has some subtlety. tf.while() will run several iterations in parallel but this is not the same as _batching_ those same iterations. For that, you'll need to use the experimental parallel_for feature [0]. Using this should get you into roughly the same speed as JAX. Tricky! [0]: https://github.com/tensorflow/tensorflow/tree/b3e007394680801113f492fa1f5a9784e8502f19/tensorflow/python/ops/parallel_for https://github.com/tensorflow/tensorflow/tree/b3e00739468080...