Repository navigation
docs: cover intruder evaluation and current feature availability - #194
Merged
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The README omits intruder detection and recent search/server features, and several examples refer to removed APIs or directories. Document the current public surface and distinguish CLI-supported scorers from API-only evaluators.
Add a complete scorer table, an explanation-free IntruderScorer example with explicit group size and default/internal semantics, explainer availability including best-of-K and iterative search, final-evaluation separation, external server configuration, sampling/window controls, the matched-mark fuzzing protocol, saved-explanation helpers, and record inspection/rerun controls.
Replace obsolete RecallScorer examples with DetectionScorer, remove unsupported tokenizer/batch_size arguments, use record.latent for filenames, correct Pipeline.run concurrency, pass a Path when saving cache splits, and point to existing example/test locations. Clarify that intervention scoring is not the existing SurprisalScorer and that intruder/embedding outputs are not CLI binary-classification summaries.
Validation: all 13 Python blocks parse; public delphi imports and local links were checked (the Offline export was checked statically because vLLM is not installed locally). The intruder and multi-scorer examples execute with an in-memory generation client and temporary outputs. The generated CLI help confirms the documented option names and choices. Whitespace checks pass. This changes only README.md; no GPU inference was run.