List your company

Jailbreak Defense on GPTFuzz, TAP, GCG, AutoDAN, and Template

24.98GPTFuzz ASRSFT
22.292440.433758.57576.7163Dec 15, 2024

Evaluation results

MethodLinks
SFTFine-tuned on=CodeAlpaca-20k2024.1224.984017.3925.7929.0427.4455.92
DAREFine-tuned on=CodeAlpaca-20k2024.1226.1642.3819.0225.528.9128.454.56
SFTFine-tuned on=GSM8K, Safety Realignment=true2024.1232.1144.7623.9136.9638.5235.2547.71
DAREFine-tuned on=GSM8K, Safety Realignment=true2024.1232.5141.4323.9136.138.5834.5148.05
SafeDecodingFine-tuned on=CodeAlpaca-20k2024.1237.6665.7136.6828.3747.8343.2539.71
RESTAFine-tuned on=GSM8K, Safety Realignment=true2024.1254.5176.1989.1386.8278.2476.985.98
RESTAdFine-tuned on=GSM8K, Safety Realignment=true2024.1255.175.7188.8684.2477.4976.286.68
IRRFine-tuned on=GSM8K, Safety Realignment=true2024.1259.5682.8685.0591.1279.6879.653.31
IRRdFine-tuned on=GSM8K, Safety Realignment=true2024.1260.3684.2986.9693.1282.182.10.86
SFTfine-tuning dataset=Alpaca Chinese2024.1266.246.1934.7864.4769.4256.2126.75
DAREfine-tuning dataset=Alpaca Chinese2024.1266.248.5734.2465.3369.8856.8426.12
RESTAFine-tuned on=CodeAlpaca-20k2024.1266.381.968.2193.1283.4778.64.36
RESTAdFine-tuned on=CodeAlpaca-20k2024.1266.882.8669.5791.6983.6378.914.05
IRRFine-tuned on=CodeAlpaca-20k2024.1271.4685.2467.9391.6989.2881.121.84
Safe LoRAfine-tuning dataset=Alpaca Chinese2024.1274.7396.1989.1365.0489.5882.930.03
Safe LoRAFine-tuned on=GSM8K, Safety Realignment=true2024.1274.7396.1989.1365.0489.5882.930.03
IRRdFine-tuned on=CodeAlpaca-20k2024.1275.1287.6271.293.1290.5983.53-0.57
Safe LORAFine-tuned on=CodeAlpaca-20k2024.1277.696.6760.8780.880.8279.353.61
SafeDecodingFine-tuned on=GSM8K, Safety Realignment=true2024.1278.3985.2492.1252.4452.4476.356.05
SafeDecodingfine-tuning dataset=Alpaca Chinese2024.1281.2766.1962.7776.2277.7972.8510.11
RESTAdfine-tuning dataset=Alpaca Chinese2024.1289.8977.6279.3599.1495.9588.39-5.43
RESTAfine-tuning dataset=Alpaca Chinese2024.1291.2878.5779.8998.8596.0188.92-5.96
IRRfine-tuning dataset=Alpaca Chinese2024.1291.5878.162.2399.4395.2385.31-2.35
IRRdfine-tuning dataset=Alpaca Chinese2024.1292.1781.962.2310095.9586.45-3.49