Linguard CAA steering vectors for Tiny Aya Earth

SafeTune extract_caa_vectors on layers 12/16/20 from the held-out calibration set (Esimuda/linguard-safety/calibration). Apply with safetune.steer.CAAModel(model, torch.load('caa_vectors.pt'), strength=0.25). Inference-time harden only (no weight changes). Metrics: https://github.com/Esimuda/pidginguard

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for Esimuda/linguard-caa-vectors

Finetuned
(17)
this model

Space using Esimuda/linguard-caa-vectors 1