List your company

Safety Evaluation on MSSbench (test)

99.66Effectiveness ScorePragma-VL
87.055290.327693.696.8724Feb 28, 2026

Evaluation results

MethodLinks
Pragma-VLBackbone=Qwen2.5-VL-7B2026.0299.6655.89
Beavertails-V harmBackbone=Qwen2.5-VL-7B2026.0299.1550.5
Beavertails-V helpBackbone=Qwen2.5-VL-7B2026.0298.9832.54
SAFE RLHF-vBackbone=Qwen2.5-VL-7B2026.0298.9852.2
Beavertails-V helpBackbone=Llava-1.5-7B2026.0298.8231.19
Beavertails-V allBackbone=Qwen2.5-VL-7B2026.0298.6545.45
DPOBackbone=Qwen2.5-VL-7B2026.0298.6553.96
Pragma-VLBackbone=Llava-1.5-7B2026.0298.6555.05
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0298.4836.53
SPA-VLBackbone=Qwen2.5-VL-7B2026.0298.4836.36
DPOBackbone=Llava-1.5-7B2026.0297.4747.89
SPA-VLBackbone=Llava-1.5-7B2026.0297.328.78
MM-RLHFBackbone=Qwen2.5-VL-7B2026.0297.1343.09
Llava-1.5-7BBackbone=Llava-1.5-7B2026.0297.1328.11
MM-RLHFBackbone=Llava-1.5-7B2026.0297.1337.03
Beavertails-V allBackbone=Llava-1.5-7B2026.0296.9743.09
SFTBackbone=Qwen2.5-VL-7B2026.0296.1345.28
SFTBackbone=Llava-1.5-7B2026.0296.1235.97
SAFE RLHF-vBackbone=Llava-1.5-7B2026.0295.9544.26
Beavertails-V harmBackbone=Llava-1.5-7B2026.0287.5440.9