Commit Graph

2 Commits

Author SHA1 Message Date
Sean Naren e952dee292
Allow string plugins (#4888)
* Allow plugin to be chosen via string

* Fix implementation, add tests

* Fix codefactor issues

* Added missing env patch

* Skip test for windows

* Reword reason

* Add skip to invalid test

* Create required_plugins function, move sharded amp requirement to plugin

* Pass AMPType, fix setter for apex

* Better doc strings

* Add exception when using apex

* Add trainer available_plugins function, warn user when plugins have been added automatically with option to override behaviour

* Fixed pep8 indent

* Fix codefactor issues

* Add env variables

* Update pytorch_lightning/cluster_environments/cluster_environment.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

* Addressed code review

* Update pytorch_lightning/plugins/plugin_connector.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

* Update pytorch_lightning/plugins/plugin_connector.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

* Update pytorch_lightning/plugins/plugin_connector.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

* Addressed more code review feedback

* Fixed docstrings

* Swapped to verbose runtime error

* Apply suggestions from code review

* Apply suggestions from code review

* Update pytorch_lightning/plugins/sharded_plugin.py

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

* Change name

* Pass trainer to plugins that may require it

* Fix sharded plugin

* Added test to ensure string sharded works

* Removed trainer typing as this breaks pep8

* Fixed doc issues

* Fixed tests

Co-authored-by: chaton <thomas@grid.ai>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
2020-12-01 20:30:49 +00:00
Sean Naren bacabaebaf
Sharded Accelerator 1/n: Expose clip gradients to plugins via abstract class (#4639)
* Added abstract precision plugin to expose clip_gradients function, use within accelerator to clip gradients

* Exclude model from override, keep optimizer (needed for sharded clip gradients), add override for O2 support apex

* Fix doc

* Applied codereview changes

* Refactored clip function to encapsulate tpu changes with tpu accelerator. Default to standard clip function for vanilla torch

* Pass correct grad clip val

* Moved var to property

* Apply code review suggestions
2020-11-12 17:18:09 +00:00