Commonsense Reasoning on Winogrande (Acc, Δ, Len, Δ%)
62.12AccuracyLCPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LCPOBackbone=DeepSeek-R1-Distill-Qwen-7B2025.08 | 62.12 | 0.95 | 375 | -64.01 | |
| Original ModelBackbone=DeepSeek-R1-Distill-Qwen-7B2025.08 | 61.17 | — | 1,042 | — | |
| LCPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.08 | 50.83 | 0.79 | 309 | -79.62 | |
| Original ModelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.08 | 50.04 | — | 1,516 | — |