Mathematical Reasoning on MGSM ALL 1.0 (test)
75.93AccuracypolyGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| polyGRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 75.93 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 73.02 | |
| LIDRBackbone=Qwen2.5-7B-Instruct2026.04 | 69.6 | |
| xRFTBackbone=Qwen2.5-7B-Instruct2026.04 | 68.47 | |
| polyGRPOBackbone=Llama3-8B-Instruct2026.04 | 68.11 | |
| MAPOBackbone=Qwen2.5-7B-Instruct2026.04 | 66.29 | |
| GRPOBackbone=Llama3-8B-Instruct2026.04 | 64.58 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 63.16 | |
| MAPOBackbone=Llama3-8B-Instruct2026.04 | 60.69 | |
| LIDRBackbone=Llama3-8B-Instruct2026.04 | 55.53 | |
| xRFTBackbone=Llama3-8B-Instruct2026.04 | 53.89 | |
| BaseBackbone=Llama3-8B-Instruct2026.04 | 52.22 |