Refactor RL model wrapper into a trainer module
#144
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
|
I am strongly in favor of this refactor |
LouisCastricato
approved these changes
Dec 20, 2022
Dahoas
approved these changes
Dec 21, 2022
| config=config, | ||
| ) | ||
| model.save_pretrained("dataset/trained_model") | ||
| trainer.save_pretrained("dataset/trained_model") |
There was a problem hiding this comment.
Do the model types we use support save_pretrained?
There was a problem hiding this comment.
Wait I don't think they do, or at least ppo doesn't. The base ppo model is just an nn.Module (not pretrained). It seems actually very annoying to save new model architectures in a huggingface format. We'll probably have to write a new config.
There was a problem hiding this comment.
Ohhhh ok wait that's weird then can we just add a save pretrained function to PPO haha
There was a problem hiding this comment.
But anyway saving doesn't have anything to do with this pr so I think it's fine for now.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
This PR refactors the RL model wrappers into "trainer" wrappers. The term "model" has semantic overloading throughout the codebase. A specific point of confusion is the
{Type}RLModels, which do not only contain models but also wrap around optimizers, schedulers, and other auxiliary data structures required for RL training.This refactor was briefly discussed in the CarperAI Discord with @cat-state. I am leaving it here as a reminder and for others to chime in.