Mathematical Reasoning on GSM8K 1319 problems (test)
96.59AccuracyMCTS + EquivPruner
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MCTS + EquivPrunerBase Model=Qwen2.5-Math-7B-Instruct, Search Strategy=Monte Carlo Tree Search + EquivPruner2025.05 | 96.59 | 18,071 | 51.89 | |
| Vanilla MCTSBase Model=Qwen2.5-Math-7B-Instruct, Search Strategy=Monte Carlo Tree Search2025.05 | 96.44 | 34,826 | 100 | |
| Vanilla MCTSBase Model=Qwen2.5-Math-1.5B-Instruct, Search Strategy=Monte Carlo Tree Search2025.05 | 91.05 | 39,337 | 100 | |
| MCTS + EquivPrunerBase Model=Qwen2.5-Math-1.5B-Instruct, Search Strategy=Monte Carlo Tree Search + EquivPruner2025.05 | 90.75 | 23,752 | 60.38 | |
| MCTS + EquivPrunerBase Model=Mistral-7b-sft, Search Strategy=Monte Carlo Tree Search + EquivPruner2025.05 | 85.06 | 12,537 | 62.01 | |
| Vanilla MCTSBase Model=Mistral-7b-sft, Search Strategy=Monte Carlo Tree Search2025.05 | 83.78 | 20,217 | 100 |