Code Generation on HumanEval (Accuracy, Tokens, Ratio)
29.27AccuracyMCTS + EquivPruner
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MCTS + EquivPrunerModel=Qwen2.5-Math-7B-Instruct, Search Strategy=MCTS + EquivPruner2025.05 | 29.27 | 737 | 96.09 | |
| Vanilla MCTSModel=Qwen2.5-Math-7B-Instruct, Search Strategy=Vanilla MCTS2025.05 | 19.51 | 767 | 100 |