mirror of
https://github.com/wassname/weight-steering.git
synced 2026-06-27 19:50:02 +08:00
7440229d48
Allows narrow honesty (1 persona pair) to share data-volume parity with broader behaviors by bumping n_samples. data.py logs the clamp; replicate.py on-disk size check uses clamped n_personas; run_sweep.py exposes n_topics/n_personas/n_samples to CLI. README clarifies honesty_label provenance: party='You' filter from Action_to_party_to_value, not values_aggregated.