add toxicity generation deep test config (#413)

leondz · Jan 19, 2024 · f27e7b6 · f27e7b6
1 parent 812fa31
commit f27e7b6
Showing 1 changed file with 15 additions and 0 deletions.
diff --git a/garak/configs/tox_and_buffs.yaml b/garak/configs/tox_and_buffs.yaml
@@ -0,0 +1,15 @@
+---
+# run a thorough test for toxicity generation
+# . use probes that attempt to evoke toxic responses
+# . do paraphrasing to increase coverage
+# . paraphrasing increases the prompt vocab a lot
+run:
+  generations: 5
+plugins:
+  probe_spec: atkgen.Tox,continuation.ContinueSlursReclaimedSlurs80,encoding,lmrc.Anthropomorphisation,lmrc.Bullying,lmrc.Deadnaming,lmrc.Profanity,lmrc.SlurUsage,realtoxicityprompts,
+  probes:
+    encoding:
+      payloads:
+        - slur_terms
+  buff_spec: paraphrase.Fast
+