Question Answering on 2WikiMultiHopQA (OOD)
2.21Exact Match (EM)MBT-S
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MBT-SModel=Qwen3 0.6B2026.02 | 2.21 | 1.69 | 1.38 | |
| MBT-RModel=Qwen3 0.6B2026.02 | 1.85 | 1.33 | 1.06 | |
| MBT-SModel=Qwen3 4B2026.02 | 1.08 | 0.82 | 0.09 | |
| MBT-SModel=Qwen3 1.7B2026.02 | 0.77 | 0.68 | 0.64 | |
| MBT-RModel=Qwen3 4B2026.02 | 0.74 | 0.47 | -0.27 | |
| GRPOModel=Qwen3 1.7B2026.02 | 0.5 | 0.38 | 0.34 | |
| GRPOModel=Qwen3 4B2026.02 | 0.43 | 0.09 | -0.63 | |
| MBT-RModel=Qwen3 1.7B2026.02 | 0.35 | 0.29 | 0.25 | |
| GRPOModel=Qwen3 0.6B2026.02 | 0.34 | -0.18 | -0.45 | |
| RSModel=Qwen3 1.7B2026.02 | 0.11 | 0.1 | 0.09 | |
| BaseModel=Qwen3 0.6B2026.02 | 0 | 0 | 0 | |
| BaseModel=Qwen3 1.7B2026.02 | 0 | 0 | 0 | |
| BaseModel=Qwen3 4B2026.02 | 0 | 0 | 0 | |
| RSModel=Qwen3 0.6B2026.02 | -0.04 | -0.02 | -0.01 | |
| PromptModel=Qwen3 4B2026.02 | -0.21 | -0.22 | -0.08 | |
| gpt-oss-distillModel=Qwen3 4B2026.02 | -0.32 | -0.67 | -1.44 | |
| PromptModel=Qwen3 0.6B2026.02 | -0.62 | -0.46 | -0.14 | |
| RSModel=Qwen3 4B2026.02 | -0.67 | -0.67 | -0.68 | |
| PromptModel=Qwen3 1.7B2026.02 | -0.91 | -0.46 | 0.23 | |
| LIMOProModel=Qwen3 0.6B2026.02 | -1.82 | -1.35 | -0.68 | |
| gpt-oss-distillModel=Qwen3 0.6B2026.02 | -2.01 | -2.52 | -2.77 | |
| TokenSkipModel=Qwen3 4B2026.02 | -2.33 | -2.37 | -2.37 | |
| gpt-oss-distillModel=Qwen3 1.7B2026.02 | -2.75 | -2.9 | -2.94 | |
| TokenSkipModel=Qwen3 0.6B2026.02 | -2.84 | -2.8 | -2.77 | |
| LIMOProModel=Qwen3 1.7B2026.02 | -3.74 | -3.05 | -2.19 | |
| TokenSkipModel=Qwen3 1.7B2026.02 | -4.48 | -4.52 | -4.53 | |
| LIMOProModel=Qwen3 4B2026.02 | -4.81 | -4.32 | -4.2 |