Question Answering on Bamboo
50.4AccuracyCoT
Evaluation Results
| Method | Links | |
|---|---|---|
| CoTBackbone=Qwen2.5-14B-Instruct2026.02 | 50.4 | |
| CDKCBackbone=Qwen2.5-14B-Instruct2026.02 | 44 | |
| GRPOBackbone=Qwen2.5-14B-Instruct2026.02 | 39.2 | |
| Vanilla SFTBackbone=Qwen2.5-14B-Instruct2026.02 | 36 | |
| CoTBackbone=Qwen2.5-3B-Instruct2026.02 | 32.8 | |
| Vanilla LLMBackbone=Qwen2.5-14B-Instruct2026.02 | 28.8 | |
| CDKCBackbone=Qwen2.5-3B-Instruct2026.02 | 26.4 | |
| RAGBackbone=Qwen2.5-14B-Instruct2026.02 | 25.6 | |
| GRPOBackbone=Qwen2.5-3B-Instruct2026.02 | 22.4 | |
| Vanilla SFTBackbone=Qwen2.5-3B-Instruct2026.02 | 20 | |
| CGKEBackbone=Qwen2.5-14B-Instruct2026.02 | 20 | |
| CGKEBackbone=Qwen2.5-3B-Instruct2026.02 | 17.6 | |
| RAGBackbone=Qwen2.5-3B-Instruct2026.02 | 13.6 | |
| Vanilla LLMBackbone=Qwen2.5-3B-Instruct2026.02 | 12 |