Visual Attribution on Thinking-Model Attribution Dataset Code
73LDSAttnLRP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AttnLRPModel=Qwen3-VL-8B-Thinking2026.04 | 73 | 0.82 | |
| TAMModel=Qwen3-VL-8B-Thinking2026.04 | 72 | 1.06 | |
| AttnLRPModel=MiMo-VL-7B2026.04 | 72 | 0.81 | |
| AttnLRPModel=GLM-4.1V-9B-Thinking2026.04 | 71 | 0.84 | |
| vStreamModel=Cosmos-Reason1-7B2026.04 | 71 | 1.04 | |
| vStreamModel=MiMo-VL-7B2026.04 | 70 | 0.96 | |
| vStreamModel=Qwen3-VL-8B-Thinking2026.04 | 69 | 0.91 | |
| AttnLRPModel=Cosmos-Reason1-7B2026.04 | 69 | 1 | |
| vStreamModel=GLM-4.1V-9B-Thinking2026.04 | 68 | 1.1 | |
| TAMModel=MiMo-VL-7B2026.04 | 68 | 1.04 | |
| TAMModel=GLM-4.1V-9B-Thinking2026.04 | 67 | 1.01 | |
| TAMModel=Cosmos-Reason1-7B2026.04 | 65 | 0.95 | |
| InputGradModel=GLM-4.1V-9B-Thinking2026.04 | 62 | 0.78 | |
| InputGradModel=Cosmos-Reason1-7B2026.04 | 61 | 0.72 | |
| InputGradModel=MiMo-VL-7B2026.04 | 59 | 0.76 | |
| InputGradModel=Qwen3-VL-8B-Thinking2026.04 | 58 | 0.96 | |
| AttentionModel=GLM-4.1V-9B-Thinking2026.04 | 45 | 0.32 | |
| AttentionModel=Cosmos-Reason1-7B2026.04 | 41 | 0.37 | |
| AttentionModel=MiMo-VL-7B2026.04 | 40 | 0.33 | |
| AttentionModel=Qwen3-VL-8B-Thinking2026.04 | 39 | 0.29 | |
| RandomModel=Qwen3-VL-8B-Thinking2026.04 | 33 | 0.09 | |
| RandomModel=GLM-4.1V-9B-Thinking2026.04 | 33 | 0.1 | |
| RandomModel=MiMo-VL-7B2026.04 | 30 | 0.16 | |
| RandomModel=Cosmos-Reason1-7B2026.04 | 30 | 0.16 |