Share your thoughts, 1 month free Claude Pro on usSee more

Multimodal Safety Evaluation on MSSBench

2.55Safety Score

Qwen2.5-VL-3B + VLGuard

Updated 1mo ago

Evaluation Results

Method	Links
Qwen2.5-VL-3B + VLGuard 2026.03		2.55	1.74	1.66
Qwen2.5-VL-7B + VLGuard 2026.03		2.48	1.83	1.88
Qwen2.5-VL-3B + TIS 2026.03		2.42	2.15	2.06
Qwen2.5-VL-3B + SFT+DPO+GRPO 2026.03		2.39	2.69	2.69
Qwen2.5-VL-3B + SaFeR-VLM 2026.03		2.32	2.21	1.93
Qwen2.5-VL-7B + SaFeR-VLM 2026.03		2.32	2.39	2.04
Qwen2.5-VL-7B + SFT+GRPO 2026.03		2.31	2.78	2.69
Qwen2.5-VL-7B + SFT+DPO+GRPO 2026.03		2.3	2.83	2.79
Qwen2.5-VL-7B + TIS 2026.03		2.26	2.46	2.26
Qwen2.5-VL-3B + SFT+GRPO 2026.03		2.19	2.58	2.55
Qwen2.5-VL-7B + SFT+DPO 2026.03		1.5	2.72	2.59
Qwen2.5-VL-7B + SIA 2026.03		1.28	2.11	1.49
Qwen2.5-VL-3B + SFT+DPO 2026.03		1.21	2.51	2.32
Qwen2.5-VL-7B + SFT 2026.03		1.11	2.6	2.27
Qwen2.5-VL-3B + SPA-VL 2026.03		0.99	1.94	1.42
Qwen2.5-VL-3B + SIA 2026.03		0.91	1.74	1.32
Qwen2.5-VL-3B + SFT 2026.03		0.76	2.32	2.03
Qwen2.5-VL-7B + SPA-VL 2026.03		0.67	2.06	1.47
Qwen2.5-VL-7B + ECSO 2026.03		0.64	2.04	1.5
Qwen2.5-VL-7B 2026.03		0.56	2.07	1.51
Qwen2.5-VL-3B + ECSO 2026.03		0.02	1.8	1.22
Qwen2.5-VL-3B 2026.03		-0.11	1.98	1.31