[bug-fix] Don't load non-wrapped policy #4593
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ervteng
changed the title
andrewcoh
approved these changes
Oct 22, 2020
dongruoping
pushed a commit
that referenced
this pull request
Oct 30, 2020
* Always initialize non-wrapped policy * Load ghosted policy * Update changelog * Resume test * Add test * Add torch test and fix torch.
dongruoping
pushed a commit
that referenced
this pull request
Oct 31, 2020
* Always initialize non-wrapped policy * Load ghosted policy * Update changelog * Resume test * Add test * Add torch test and fix torch.
10 tasks
dongruoping
pushed a commit
that referenced
this pull request
Nov 2, 2020
* [bug-fix] Don't load non-wrapped policy (#4593) * pin cattrs version * cap PyTorch version * use v2 action and pin python version (#4568) Co-authored-by: Ervin T <[email protected]> Co-authored-by: Chris Elion <[email protected]>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Proposed change(s)
The ghost trainer creates an unwrapped policy for ghosting (that doesn't belong to any trainer), and initializes/loads it. This causes an issue on load for TensorFlow, since the saving ops are not added to the policy (normally they are added when the policy is added to a trainer). However, the checkpoint does contain these ops, so when we attempt to load it, TF throws an error.
This change makes sure the unwrapped policy is never loaded. This policy will then receive the weights of the wrapped policy, so it's not an issue.
Types of change(s)
Checklist
Other comments