# LSTM/RNN documentation

**URL:** <https://discuss.ray.io/t/lstm-rnn-documentation/170>\
**Category:** RLlib\
**Created:** [December 8, 2020, 11:36pm UTC](https://discuss.ray.io/t/lstm-rnn-documentation/170 "2020-12-08T23:36:00Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Jeffrey\_Waldron](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/jeffrey_waldron/32/36_2.png) [@Jeffrey\_Waldron](https://discuss.ray.io/u/Jeffrey_Waldron)\
**Post date:** [December 8, 2020, 11:36pm UTC](https://discuss.ray.io/t/lstm-rnn-documentation/170/1 "2020-12-08T23:36:00Z")

</div>

It would be great to have a more complete description of how RLLib handles the observation pipeline to feed LSTM/RNN models, and is there any difference between the provided option and a custom implementation. Particularly how it changes what a typical batch looks like, how does `max_seq_len` value impact the pipeline, and should it change your choice of batch mode (“complete\_episodes”??).