Visual Attribution on Thinking-Model Attribution Dataset Math
76LDSAttnLRP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AttnLRPModel=Qwen3-VL-8B-Thinking2026.04 | 76 | 0.81 | |
| AttnLRPModel=Cosmos-Reason1-7B2026.04 | 76 | 0.8 | |
| vStreamModel=Qwen3-VL-8B-Thinking2026.04 | 75 | 1.05 | |
| vStreamModel=MiMo-VL-7B2026.04 | 75 | 1.08 | |
| TAMModel=Cosmos-Reason1-7B2026.04 | 75 | 0.94 | |
| AttnLRPModel=MiMo-VL-7B2026.04 | 74 | 0.8 | |
| vStreamModel=Cosmos-Reason1-7B2026.04 | 74 | 0.99 | |
| AttnLRPModel=GLM-4.1V-9B-Thinking2026.04 | 72 | 0.79 | |
| TAMModel=GLM-4.1V-9B-Thinking2026.04 | 71 | 0.94 | |
| TAMModel=MiMo-VL-7B2026.04 | 70 | 0.98 | |
| TAMModel=Qwen3-VL-8B-Thinking2026.04 | 69 | 1.02 | |
| vStreamModel=GLM-4.1V-9B-Thinking2026.04 | 69 | 0.97 | |
| InputGradModel=Qwen3-VL-8B-Thinking2026.04 | 63 | 0.72 | |
| InputGradModel=GLM-4.1V-9B-Thinking2026.04 | 58 | 0.74 | |
| InputGradModel=MiMo-VL-7B2026.04 | 58 | 1.06 | |
| InputGradModel=Cosmos-Reason1-7B2026.04 | 58 | 0.71 | |
| AttentionModel=MiMo-VL-7B2026.04 | 45 | 0.34 | |
| AttentionModel=Cosmos-Reason1-7B2026.04 | 45 | 0.33 | |
| AttentionModel=GLM-4.1V-9B-Thinking2026.04 | 41 | 0.3 | |
| AttentionModel=Qwen3-VL-8B-Thinking2026.04 | 39 | 0.31 | |
| RandomModel=Qwen3-VL-8B-Thinking2026.04 | 31 | 0.08 | |
| RandomModel=MiMo-VL-7B2026.04 | 30 | 0.14 | |
| RandomModel=GLM-4.1V-9B-Thinking2026.04 | 28 | 0.12 | |
| RandomModel=Cosmos-Reason1-7B2026.04 | 28 | 0.17 |