You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jan 15, 2024. It is now read-only.
apache/mxnet#11948 introduces foreach to the contrib.unroll method which makes RNNs hybridizable and easy to export to json for deployment. In theory it introduces extra overhead such as calling nd.where. We need to evaluate the performance impact (I assume it's relatively small) and change gluon nlp language models to hybrid blocks for ease of deployment if possible. It should not affect existing pre-trained models since the parameter structure won't change.
szha
changed the title
[Discussion] Use foreach to unroll all language models
[Discussion] Use control-flow operators (e.g. foreach) to unroll dynamic graph models
Aug 28, 2018
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
apache/mxnet#11948 introduces foreach to the contrib.unroll method which makes RNNs hybridizable and easy to export to json for deployment. In theory it introduces extra overhead such as calling
nd.where
. We need to evaluate the performance impact (I assume it's relatively small) and change gluon nlp language models to hybrid blocks for ease of deployment if possible. It should not affect existing pre-trained models since the parameter structure won't change.@zheng-da @szha
The text was updated successfully, but these errors were encountered: