General Tasks on GPQA Diamond
0.677PAQ ScoreUCPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| UCPOBackbone Model=Qwen3-8B2026.01 | 0.677 | 0.5616 | |
| GRPO-UCBackbone Model=Qwen3-8B, Reward Coefficient (ru)=0.52026.01 | 0.6509 | 0.5865 | |
| GRPO-UCBackbone Model=Qwen3-8B, Reward Coefficient (ru)=0.82026.01 | 0.64 | 0.6005 | |
| GRPO-UCBackbone Model=Qwen3-8B, Reward Coefficient (ru)=0.22026.01 | 0.6025 | 0.5906 | |
| GRPOBackbone Model=Qwen3-8B2026.01 | 0.5935 | 0.5879 | |
| Prompt-UCBackbone Model=Qwen3-8B2026.01 | 0.5684 | 0.5567 | |
| BaselineBackbone Model=Qwen3-8B2026.01 | 0.5657 | 0.5657 | |
| UCPOBackbone Model=Llama-3.1-8B-Instruct2026.01 | 0.3602 | 0.1718 | |
| GRPO-UCBackbone Model=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.82026.01 | 0.3432 | 0.1952 | |
| GRPO-UCBackbone Model=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.52026.01 | 0.3421 | 0.1898 | |
| GRPO-UCBackbone Model=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.22026.01 | 0.2946 | 0.2946 | |
| GRPOBackbone Model=Llama-3.1-8B-Instruct2026.01 | 0.2727 | 0.2727 | |
| Prompt-UCBackbone Model=Llama-3.1-8B-Instruct2026.01 | 0.2328 | 0.2274 | |
| BaselineBackbone Model=Llama-3.1-8B-Instruct2026.01 | 0.2256 | 0.2256 |