Counterfactual Prediction on CausalVLBench
87.7PrecisionBRIDGEVLM-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BRIDGEVLM-7BModel Size=7B2026.06 | 87.7 | 90 | 83.7 | 98.5 | |
| QWEN2.5-VL-32BModel Size=32B2026.06 | 87.4 | 90.8 | 86.7 | 98.4 | |
| GEMINI-2.0-FLASH (CM)Category=Commercial2026.06 | 86.5 | 90.7 | 88.3 | 97.4 | |
| LLAVA-ONEVISION-7BModel Size=7B2026.06 | 83.5 | 88.5 | 85 | 96.9 | |
| DEEPSEEK-VL2-S-16BModel Size=16B2026.06 | 77.9 | 57 | 51.4 | 41.6 | |
| PHI-4-MMI-7B (CAUSAL-TRACE)Model Size=7B, Setting=Causal-trace supervision2026.06 | 76.5 | 84.8 | 83.1 | 94.9 | |
| PHI-4-MMI-7B (ANSWER)Model Size=7B, Setting=Answer-only fine-tuning2026.06 | 76.2 | 84.2 | 80.9 | 95.6 | |
| PHI-4-MMI-7B (NO-FT, CAUSAL-PROMPT)Model Size=7B, Setting=No-FT, Causal-Prompt2026.06 | 63 | 76.8 | 73.3 | 94 |