From 611ed24fcab76184a294bb747d83a94db3912d75 Mon Sep 17 00:00:00 2001 From: Prannaya Date: Sun, 4 Aug 2024 00:25:33 +0800 Subject: [PATCH] feat(experiments): final mutation generation script (tested on mistral 7B with both datasets) --- experiments/mutation_generation.py | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/experiments/mutation_generation.py b/experiments/mutation_generation.py index a74d706..730c2c0 100644 --- a/experiments/mutation_generation.py +++ b/experiments/mutation_generation.py @@ -26,7 +26,7 @@ dataset_args = { "harmbench": ("walledai/HarmBench", "standard"), "advbench": ("walledai/AdvBench", ), - "xstest": ("walledai/XSTest"), + "xstest": ("walledai/XSTest", ), } if __name__ == "__main__": @@ -37,7 +37,7 @@ help="Model to use") parser.add_argument("-d", "--dataset", - default="harmbench", choices=["harmbench", "advbench"], help="(Prompt-based) Dataset to test") + default="harmbench", choices=["harmbench", "advbench", "xstest"], help="(Prompt-based) Dataset to test") parser.add_argument("-f", "--filename", default="", help="Place to store logs") @@ -52,7 +52,7 @@ help="Print running logs", action="store_true") parser.add_argument("-i", "--interval", - type=int, default=1, + type=int, default=100, help="Number of runs before saving") parser.add_argument("-n", "--num", @@ -115,7 +115,7 @@ tactics = [ #"autodan/revise", - "masterkey/rephrase", + #"masterkey/rephrase", "renellm/alter-sentence-structure", "renellm/change-style", "renellm/insert-meaningless-characters",