Commonsense Reasoning on X-CSQA EN 1.0 (test)
82AccuracypolyGRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| polyGRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 82 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 77.1 | |
| MAPOBackbone=Qwen2.5-7B-Instruct2026.04 | 77.1 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.04 | 75.5 | |
| LIDRBackbone=Qwen2.5-7B-Instruct2026.04 | 75.1 | |
| polyGRPOBackbone=Llama3-8B-Instruct2026.04 | 70.9 | |
| xRFTBackbone=Qwen2.5-7B-Instruct2026.04 | 70.3 | |
| LIDRBackbone=Llama3-8B-Instruct2026.04 | 69.5 | |
| GRPOBackbone=Llama3-8B-Instruct2026.04 | 68.8 | |
| xRFTBackbone=Llama3-8B-Instruct2026.04 | 67.7 | |
| BaseBackbone=Llama3-8B-Instruct2026.04 | 66.3 | |
| MAPOBackbone=Llama3-8B-Instruct2026.04 | 62.3 |