Upload nemotron-4B/aegis-test-ru/probes_binary_eval/test_summary.txt with huggingface_hub

nemotron-4B/aegis-test-ru/probes_binary_eval/test_summary.txt

@@ -0,0 +1,31 @@
+Binary probe TEST evaluation (1961 samples)
+Probe trained on train split, best layer: 20
+Label column: primary_cat, positive value: != safe
+
+Class distribution (test):
+  safe:   719
+  unsafe: 1242
+
+============================================================
+Accuracy:  0.6369
+F1:        0.7768
+F1-macro:  0.4021
+ROC-AUC:   0.8042
+PR-AUC:    0.8719
+============================================================
+
+Bootstrap 95% CI (n=1000):
+  Accuracy:  [0.6145, 0.6573]
+  F1:        [0.7594, 0.7922]
+  F1-macro:  [0.3898, 0.4148]
+  ROC-AUC:   [0.7853, 0.8226]
+  PR-AUC:    [0.8524, 0.8903]
+
+              precision    recall  f1-score   support
+
+        safe       0.77      0.01      0.03       719
+      unsafe       0.64      1.00      0.78      1242
+
+    accuracy                           0.64      1961
+   macro avg       0.70      0.51      0.40      1961
+weighted avg       0.68      0.64      0.50      1961