Jailbreak Attack on JBB (Multi-Model ASR/Loss Evaluation) benchmark leaderboard · SOTA2 Research

Jailbreak Attack on JBB (Multi-Model ASR/Loss Evaluation)

91.7Llama2-7B ASRFaster-I-GCG

Evaluation Results

Method	Links
Faster-I-GCG2024.10		91.7	0.106	1.201	48.4	1.352	76.3	—	—	—	—	—
I-GCG2024.10		89.6	0.108	1.453	43.1	1.315	77.1	1.169	56.1	0.349	58.2	64.8
Faster-GCG++2024.10		71.4	0.17	0.653	77.6	0.348	84.2	0.466	68.6	0.149	88.7	78.1
TAO2024.10		68.9	0.278	0.95	50.1	0.44	83.6	0.672	56.6	0.296	74.5	66.7
Faster-GCG2024.10		66	0.268	0.863	54.2	0.445	81.4	0.514	67.1	0.251	78.7	69.5
GCG2024.10		58.2	0.35	0.981	47.7	0.485	80.6	0.639	64.5	0.312	77.9	65.8
GCG-PS2024.10		49.4	0.531	1.776	29.4	0.661	71.8	1.168	41.5	—	—	—
PAIR2024.10		42.9	—	—	65.3	—	96.8	—	85	—	47	67.4
FLIP2024.10		35	—	—	37.8	—	88	—	57.5	—	49.2	53.5
AutoDAN2024.10		27.7	1.825	4.341	23.1	5.74	71.8	1.854	36.1	—	—	—
PAP2024.10		26.9	—	—	49.7	—	90.5	—	89.7	—	42.8	59.9
TRIAL2024.10		24.3	—	—	28.6	—	93.1	—	84.9	—	7.5	47.7