Please make sure that this is a documentation issue. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:doc_template

System information

• TensorFlow version: TF r1.13 (and many other versions)
• Doc Link: tf.nn.dynamic_rnn

Describe the documentation issue

It describes the argument sequence_length as "it's more for performance than correctness", but this is wrong:

• This argument enables this API to extract the last VALID state of RNN instead of a PADDED time step, so it IS for correctness.
• This argument CANNOT reduce computation, because even with this argument, computation of new states of RNN cell still occurs. The only difference is that tf.nn.dynamic_rnn chooses to copy through old states instead of keeping the new states, and this leads to even larger computation. So it is NOT for performance.

The initial document is correct, see this commit.

it's more for correctness than performance"

I don't really understand why you replaced the correct document with a wrong one.

We welcome contributions by users. Will you be able to update submit a PR (use the doc style guide) to fix the doc Issue?