Context Learning on CL-Bench (test)
12.85Overall ScoreContext-CoT SFT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Context-CoT SFTmodel=Qwen3.5-4B2026.05 | 12.85 | 3.79 | 14.88 | 12.31 | 10.25 | 12.72 | |
| Answer-only SFTmodel=Qwen3.5-4B2026.05 | 9.32 | 0.26 | 10.87 | 8.38 | 7.66 | 9.14 | |
| Base Modelmodel=Qwen3.5-4B2026.05 | 9.06 | — | 9.63 | 11.05 | 6.91 | 9.4 | |
| Answer-exposed CoT SFTmodel=Qwen3.5-4B2026.05 | 8.59 | -0.47 | 9.66 | 9.52 | 7.23 | 8.08 | |
| Context-CoT SFTmodel=Llama3.2-3B2026.05 | 7.57 | 4.53 | 9.11 | 5.76 | 7.89 | 6.2 | |
| Answer-only SFTmodel=Llama3.2-3B2026.05 | 4.25 | 1.21 | 4.19 | 5.18 | 2.91 | 5.02 | |
| Base Modelmodel=Llama3.2-3B2026.05 | 3.04 | — | 3.44 | 2.59 | 3.29 | 2.52 | |
| Answer-exposed CoT SFTmodel=Llama3.2-3B2026.05 | 2.71 | -0.33 | 2.22 | 2.21 | 3.19 | 3.08 |