Intervention Target Prediction on Causal3D
92.3Average AccuracyBRIDGEVLM-7B
Evaluation results
| Method | Links | |
|---|---|---|
| BRIDGEVLM-7BModel Size=7B2026.06 | 92.3 | |
| PHI-4-MMI-7B (ANSWER)Model Size=7B, Setting=Answer-only fine-tuning2026.06 | 81.6 | |
| PHI-4-MMI-7B (CAUSAL-TRACE)Model Size=7B, Setting=Causal-trace supervision2026.06 | 81 | |
| PHI-4-MMI-7B (NO-FT, CAUSAL-PROMPT)Model Size=7B, Setting=No-FT, Causal-Prompt2026.06 | 77.6 | |
| GEMINI-2.0-FLASH (CM)Category=Commercial2026.06 | 65 | |
| LLAVA-ONEVISION-7BModel Size=7B2026.06 | 60.1 | |
| DEEPSEEK-VL2-S-16BModel Size=16B2026.06 | 53 | |
| BRIDGEVLM-7BModel Size=7B2026.06 | 49 | |
| PHI-4-MMI-7B (CAUSAL-TRACE)Model Size=7B, Setting=Causal-trace supervision2026.06 | 43.6 | |
| PHI-4-MMI-7B (ANSWER)Model Size=7B, Setting=Answer-only fine-tuning2026.06 | 43.3 | |
| LLAVA-ONEVISION-7BModel Size=7B2026.06 | 39.8 | |
| PHI-4-MMI-7B (NO-FT, CAUSAL-PROMPT)Model Size=7B, Setting=No-FT, Causal-Prompt2026.06 | 38.7 | |
| DEEPSEEK-VL2-S-16BModel Size=16B2026.06 | 33.8 | |
| GEMINI-2.0-FLASH (CM)Category=Commercial2026.06 | 33.5 |