Mathematical Reasoning on NuminaMath in-distribution LEAN (test)
65.17Average@32GSPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GSPOModel=Qwen, Checkpoint Selection=best checkpoint2026.06 | 65.17 | 65.17 | |
| GRPOModel=Qwen, Checkpoint Selection=best checkpoint2026.06 | 64.83 | 64.83 | |
| DAPOModel=Qwen, Checkpoint Selection=best checkpoint2026.06 | 63.02 | 63.02 | |
| WAPOModel=Qwen, Checkpoint Selection=best checkpoint2026.06 | 61.99 | 61.99 | |
| DAPOModel=Smol, Checkpoint Selection=best checkpoint2026.06 | 58.79 | 58.79 | |
| GSPOModel=Smol, Checkpoint Selection=best checkpoint2026.06 | 58.4 | 58.4 | |
| GRPOModel=Smol, Checkpoint Selection=best checkpoint2026.06 | 58.02 | 58.02 | |
| WAPOModel=Smol, Checkpoint Selection=best checkpoint2026.06 | 57.63 | 57.63 | |
| GSPOModel=Gemma, Checkpoint Selection=best checkpoint2026.06 | 44.27 | 44.27 | |
| GRPOModel=Gemma, Checkpoint Selection=best checkpoint2026.06 | 43.39 | 43.39 | |
| WAPOModel=Gemma, Checkpoint Selection=best checkpoint2026.06 | 42.78 | 42.78 | |
| DAPOModel=Gemma, Checkpoint Selection=best checkpoint2026.06 | 42.65 | 42.65 |