The model in test_machine_translation.py is not correct.

[lcy-seso]

This script is expected to implement the encoder-decoder without attention. , but it is not the conventional RNN encoder-decoder. See this paper for reference https://arxiv.org/pdf/1406.1078.pdf

In every time step, decoder takes the encoded vector as its input: https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/fluid/tests/book/test_machine_translation.py#L36

Also related to #7011