Math and Text Reasoning on Minerva
0.4943PAQPrompt-UC
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Prompt-UCBackbone=Qwen3-8B2026.01 | 0.4943 | 0.486 | |
| UCPOBackbone=Qwen3-8B2026.01 | 0.4915 | 0.4763 | |
| GRPO-UCBackbone=Qwen3-8B, Reward Coefficient (ru)=0.22026.01 | 0.486 | 0.4762 | |
| GRPO-UCBackbone=Qwen3-8B, Reward Coefficient (ru)=0.82026.01 | 0.4755 | 0.4693 | |
| GRPOBackbone=Qwen3-8B2026.01 | 0.4718 | 0.4662 | |
| GRPO-UCBackbone=Qwen3-8B, Reward Coefficient (ru)=0.52026.01 | 0.4632 | 0.4551 | |
| BaselineBackbone=Qwen3-8B2026.01 | 0.4596 | 0.4596 | |
| GRPO-UCBackbone=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.52026.01 | 0.2616 | 0.0911 | |
| GRPO-UCBackbone=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.82026.01 | 0.25 | 0.0487 | |
| UCPOBackbone=Llama-3.1-8B-Instruct2026.01 | 0.225 | 0.1848 | |
| GRPO-UCBackbone=Llama-3.1-8B-Instruct, Reward Coefficient (ru)=0.22026.01 | 0.1988 | 0.115 | |
| GRPOBackbone=Llama-3.1-8B-Instruct2026.01 | 0.1816 | 0.1815 | |
| Prompt-UCBackbone=Llama-3.1-8B-Instruct2026.01 | 0.1654 | 0.1604 | |
| BaselineBackbone=Llama-3.1-8B-Instruct2026.01 | 0.1581 | 0.1581 |