[Reward] initial CLoud Reward trainer #2432
Open
Conversation
This file contains 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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
qgallouedec
reviewed
Dec 8, 2024
|
|
||
| [Critique-out-Loud reward models](https:/huggingface.co/papers/2408.11791) are reward models that can reason explicitly about the quality of an input through producing Chain-of-Thought like critiques of an input before predicting a reward. In classic reward model training, the reward model is trained as a reward head initialized on top of the base LLM. Without LM capabilities, classic reward models act as encoders and must predict rewards within a single forward pass through the model, meaning reasoning must happen implicitly. In contrast, CLoud reward models are trained to both produce explicit reasoning about quality and to score based on these critique reasoning traces. | ||
|
|
||
| To train a Critique-out-Loud reward model, you can use the `feedback_method="teacher"` and set the `lm_weight` to a high value. |
There was a problem hiding this comment.
How to set the lm_height to a high value? Maybe add a ref or a code example
|
|
||
| To train a Critique-out-Loud reward model, you can use the `feedback_method="teacher"` and set the `lm_weight` to a high value. | ||
|
|
||
| The dataset should contain the columns `"prompt"`, `"chosen"`, `"rejected"` as well as the `"chosen_feedback` and `"rejected_feedback"` columns which contain the Chain-of-Thought like critiques for the chosen and rejected responses respectively generated by the same model or a different model. A script to generate such a dataset from a dataset of preferences is available in the `examples/datasets/critique_out_loud_vllm.py` file. |
| def test_train_with_feedback(self): | ||
| with tempfile.TemporaryDirectory() as tmp_dir: | ||
| # Create a dummy dataset with feedback | ||
| dummy_dataset_dict = { |
There was a problem hiding this comment.
Ok now I get why you want a tiny version: the zen dataset misses two columns.
I'll think about the best way to do it
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.
What does this PR do?
Adds an option to the
RewardTrainerto implement the CLoud method.