2 days ago
-
analysis (full)
-
results 12/12 (full) (model replies, private store only)
-
results 12/12 (full)
-
held-out sets built
3 days ago
-
analysis (full)
-
results 12/12 (full) (model replies, private store only)
-
results 12/12 (full)
-
held-out sets built
4 days ago
-
export manifest (models for the detailed-evaluation notebook)
-
AuditKIT independent safety scoring
-
model replies for AuditKIT scoring (private store only)
-
paired differences vs FP16 + forest plot
-
final analysis
-
eval results 26/26
-
eval results 26/26
-
CircuitKIT ranking + CK arms added
-
Upload scores/ck_scores_sidecar.json with huggingface_hub
-
CircuitKIT layer scores (HF Job)
-
before CircuitKIT HF job (patch_groups.json + first_tokens.json must be in the store)
-
consensus + size-ablation arms added
5 days ago
-
Add README
-
Create results_table.csv
-
AuditKIT independent safety scoring
-
model replies for AuditKIT scoring (private store only)
-
final analysis
-
eval results 20/20
-
gradient attribution layer scores + arms
-
final analysis
-
eval results 18/18
-
frozen experiment grid
-
layer scoring findings + all scores
-
initial commit