Reasoning failure prediction and recovery on CRUXEval L2
77Accuracythought-tree-based classifier
Evaluation results
| Method | Links | ||
|---|---|---|---|
| thought-tree-based classifierTarget Reasoning Model=QwQ, Evaluation Protocol=Generalization from L1 training2026.04 | 77 | 13 | |
| thought-tree-based classifierTarget Reasoning Model=R1, Evaluation Protocol=Generalization from L1 training2026.04 | 75 | 22 | |
| thought-tree-based classifierTarget Reasoning Model=R1, Evaluation Protocol=Generalization from L1 training2026.04 | 0.83 | 0.21 | |
| thought-tree-based classifierTarget Reasoning Model=QwQ, Evaluation Protocol=Generalization from L1 training2026.04 | 0.74 | 0.06 |