List your company

Safety Evaluation on SPA-VL (test)

97.93Helpfulness ScorePragma-VL
17.205238.162659.1280.0774Feb 28, 2026

Evaluation results

MethodLinks
Pragma-VLBackbone=Llava-1.5-7B2026.0297.9392.05
SFTBackbone=Llava-1.5-7B2026.0289.9186.79
Pragma-VLBackbone=Qwen2.5-VL-7B2026.0287.1787.92
Beavertails-V helpBackbone=Llava-1.5-7B2026.0280.2757.74
SPA-VLBackbone=Llava-1.5-7B2026.0273.386.79
DPOBackbone=Llava-1.5-7B2026.0269.3393.96
SAFE RLHF-vBackbone=Llava-1.5-7B2026.0268.4878.87
SFTBackbone=Qwen2.5-VL-7B2026.0263.6464.91
MM-RLHFBackbone=Llava-1.5-7B2026.0262.566.79
Beavertails-V allBackbone=Llava-1.5-7B2026.0261.5165.28
Beavertails-V harmBackbone=Llava-1.5-7B2026.0256.781.13
DPOBackbone=Qwen2.5-VL-7B2026.0252.4778.87
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.025050
Llava-1.5-7BBackbone=Llava-1.5-7B2026.025050
SAFE RLHF-vBackbone=Qwen2.5-VL-7B2026.0245.0861.51
Beavertails-V helpBackbone=Qwen2.5-VL-7B2026.0240.4729.43
Beavertails-V allBackbone=Qwen2.5-VL-7B2026.0237.7151.69
Beavertails-V harmBackbone=Qwen2.5-VL-7B2026.0226.8845.66
SPA-VLBackbone=Qwen2.5-VL-7B2026.0226.7946.04
MM-RLHFBackbone=Qwen2.5-VL-7B2026.0220.3148.09