Countdown on Countdown (test)
57.7AccuracyRLVR (with verifier)
Evaluation Results
| Method | Links | |
|---|---|---|
| RLVR (with verifier)Reasoning budget=2048 tokens, Model size=1.5B2025.11 | 57.7 | |
| RAROReasoning budget=2048 tokens, Model size=1.5B2025.11 | 54.4 | |
| SFTReasoning budget=2048 tokens, Model size=1.5B2025.11 | 40.7 | |
| Iterative DPOReasoning budget=2048 tokens, Model size=1.5B2025.11 | 40.4 | |
| RationalizationReasoning budget=2048 tokens, Model size=1.5B2025.11 | 12.5 | |
| RL-LogitReasoning budget=2048 tokens, Model size=1.5B2025.11 | 2.2 | |
| BaseReasoning budget=2048 tokens, Model size=1.5B2025.11 | 2 |