bamdadd/steerbench

Document the CSV data contract and the repeng consume-not-extract boundary 1 day ago

documentation good first issue

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Property-based tests for the MMLU / GSM8K answer extractors 1 day ago

good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Make the report effect column selectable (report.py hardcodes 'formality') 1 day ago

enhancement good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Clear error for an out-of-range layer in normalize_alpha / dose 16 days ago

enhancement good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Error when --side-csv is named but missing (don't silently stub) 16 days ago

enhancement good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Validate --stem is non-empty in steer-report 16 days ago

enhancement good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Add a --json output flag to steer-report 22 days ago

help wanted good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Validate slice size (n >= 1) in the MMLU / GSM8K side-effect loaders 22 days ago

help wanted good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

load_vector: clear errors for empty or direction-less vector files 22 days ago

help wanted good first issue

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Fix matplotlib typing in report.py so mypy is clean under the report extra 23 days ago

good first issue type-safety

bamdadd/steerbench

0

The steering-vector report card: effect size, side effects, dose-response, and per-layer sensitivity for concept vectors.

Python

Open Issues Need Help