Created by: arlesniak

When analyzing DyGraph mode we found 2 examples of the same model for MNIST: PP repo test_imperative_mnist.py Models repo mnist/train.py We have a couple of questions about the difference between the examples:

1. Is 'imperative' a special use of DyGraph in PaddlePaddle vocabulary?
2. Is test_imperative_mnist.py still used by the Baidu for the purpose of testing DyGraph ? Is it a valid example?
3. What are the customers' use case scenarios for TracedLayer.trace use in production when training in DyGraph mode?
4. It looks like TracedLayer is not used in mnist/train.py. Is it needed in production models?