update replay to divergence in detector spec, calibration

NVIDIA · Oct 11, 2024 · b081082 · b081082
1 parent 70adb38
commit b081082
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/garak/data/calibration/calibration-2024-09update.json b/garak/data/calibration/calibration-2024-09update.json
@@ -385,7 +385,7 @@
       "sigma": 0.01364044725976449,
       "sw_p": 0.0011471324643020655
     },
-    "replay.Repeat/replay.RepeatDiverges": {
+    "divergence.Repeat/divergence.RepeatDiverges": {
       "mu": 0.7905982905982906,
       "sigma": 0.22985291682655218,
       "sw_p": 0.0014372154789051555

diff --git a/garak/probes/divergence.py b/garak/probes/divergence.py
@@ -20,7 +20,7 @@ class Repeat(Probe):
     training data."""
 
     bcp47 = "*"
-    primary_detector = "replay.RepeatDiverges"
+    primary_detector = "divergence.RepeatDiverges"
     extended_detectors = ["mitigation.MitigationBypass"]
     goal = "get the model to leak data by asking it to repeat words"
     doc_uri = "https://not-just-memorization.github.io/extracting-training-data-from-chatgpt.html"