Toxicity Generation on VAJA (test)
17.9Toxicity Score (Perspective API)ZO-SPSA
Evaluation results
| Method | Links | |
|---|---|---|
| ZO-SPSASurrogate Model=MiniGPT-4, Surrogate Backbone=Vicuna, Target Model=LLaVA, Target Backbone=LLaMA-2-Chat2026.01 | 17.9 | |
| ZO-SPSASurrogate Model=InstructBLIP, Surrogate Backbone=Vicuna, Target Model=LLaVA, Target Backbone=LLaMA-2-Chat2026.01 | 20.6 | |
| ZO-SPSASurrogate Model=LLaVA, Surrogate Backbone=LLaMA-2-Chat, Target Model=MiniGPT-4, Target Backbone=Vicuna2026.01 | 44.8 | |
| ZO-SPSASurrogate Model=LLaVA, Surrogate Backbone=LLaMA-2-Chat, Target Model=InstructBLIP, Target Backbone=Vicuna2026.01 | 46.5 | |
| ZO-SPSASurrogate Model=LLaVA, Surrogate Backbone=LLaMA-2-Chat, Target Model=LLaVA, Target Backbone=LLaMA-2-Chat2026.01 | 52.3 | |
| ZO-SPSASurrogate Model=InstructBLIP, Surrogate Backbone=Vicuna, Target Model=MiniGPT-4, Target Backbone=Vicuna2026.01 | 52.4 | |
| ZO-SPSASurrogate Model=MiniGPT-4, Surrogate Backbone=Vicuna, Target Model=InstructBLIP, Target Backbone=Vicuna2026.01 | 57.5 | |
| ZO-SPSASurrogate Model=InstructBLIP, Surrogate Backbone=Vicuna, Target Model=InstructBLIP, Target Backbone=Vicuna2026.01 | 61.3 | |
| ZO-SPSASurrogate Model=MiniGPT-4, Surrogate Backbone=Vicuna, Target Model=MiniGPT-4, Target Backbone=Vicuna2026.01 | 67.2 |