Mathematical Reasoning on MGSM EN 1.0 (test)
94AccuracyxRFT
Evaluation Results
| Method | Links | |
|---|---|---|
| xRFTBackbone=Qwen2.5-7B-Instruct2026.04 | 94 | |
| polyGRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 94 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 90.4 | |
| LIDRBackbone=Qwen2.5-7B-Instruct2026.04 | 90 | |
| MAPOBackbone=Qwen2.5-7B-Instruct2026.04 | 84.8 | |
| polyGRPOBackbone=Llama3-8B-Instruct2026.04 | 82 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 81.2 | |
| GRPOBackbone=Llama3-8B-Instruct2026.04 | 80.8 | |
| MAPOBackbone=Llama3-8B-Instruct2026.04 | 80.4 | |
| BaseBackbone=Llama3-8B-Instruct2026.04 | 79.6 | |
| LIDRBackbone=Llama3-8B-Instruct2026.04 | 79.6 | |
| xRFTBackbone=Llama3-8B-Instruct2026.04 | 73.2 |