Failure Detection and Reasoning on ARMBench
65Detect Acc.Claude-3.7
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Claude-3.7shots=32026.02 | 65 | 68.5 | 72.5 | |
| SFT-D2026.02 | 64 | 60.9 | 61.8 | |
| Claude-3.7shots=02026.02 | 59 | 49.4 | 38.7 | |
| Qwen2.5-VLshots=02026.02 | 50 | 34.9 | 20.8 | |
| LLaVA-NeXT2026.02 | 50 | 0 | 6.7 | |
| Cosmos-Reasoning2026.02 | 48 | 50.3 | 48 |