Cleaning up oddities with steering vecs and repe algo #72
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Outdated suggestions cannot be applied.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 cleans up some TODOs from the previous PR #66 pointed out by @dtch1997. Specifically, this PR makes 2 changes:
train_steering_vector()
function takes in a param namedread_token_index
, which can be used to read a token other than the final token. We use the second-to-last token for CAA.PipelineHook
used to control the pipeline and apply steering vectors is now a class, where all relevant state to control how the steering vector gets applied is stored, along with the steering vector itself. This should be a lot more understandable to work with and edit, and removes the weirdness where the hook is closing over variables in the original repe algorithm class.