Commonsense Question Answering on CommonsenseQA (out-of-domain)
70.7AccuracyS2L-PO-4B
Evaluation Results
| Method | Links | |
|---|---|---|
| S2L-PO-4BBase Model=Qwen3-14B-Base2026.05 | 70.7 | |
| S2L-PO-4BBase Model=Qwen3-8B-Base2026.05 | 67.8 | |
| GRPOBase Model=Qwen3-14B-Base2026.05 | 67.2 | |
| S2L-PO-1.7BBase Model=Qwen3-8B-Base2026.05 | 64.2 | |
| GRPOBase Model=Qwen3-8B-Base2026.05 | 63.9 |