Long-context Generation (Reasoning) on AIME24
46.2TPTKnapSpec
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| KnapSpecBackbone=Qwen3-4B, Layers (L)=362026.02 | 46.2 | 1.36 | 87.9 | |
| DELBackbone=Qwen3-4B, Layers (L)=362026.02 | 40.77 | 0.98 | 44.9 | |
| ARBackbone=Qwen3-4B, Layers (L)=362026.02 | 34.78 | 1 | — | |
| KnapSpecBackbone=Qwen3-8B, Layers (L)=362026.02 | 33.47 | 1.28 | 89.8 | |
| KnapSpecBackbone=Qwen3-32B, Layers (L)=642026.02 | 31.06 | 1.43 | 93.5 | |
| KnapSpecBackbone=Qwen3-14B, Layers (L)=402026.02 | 30.53 | 1.3 | 86.1 | |
| CLaSpBackbone=Qwen3-4B, Layers (L)=362026.02 | 29.68 | 1.15 | 90.9 | |
| SWIFTBackbone=Qwen3-4B, Layers (L)=362026.02 | 27.54 | 0.88 | 80.3 | |
| ARBackbone=Qwen3-8B, Layers (L)=362026.02 | 25.15 | 1 | — | |
| DELBackbone=Qwen3-8B, Layers (L)=362026.02 | 22.96 | 0.82 | 0 | |
| DELBackbone=Qwen3-14B, Layers (L)=402026.02 | 22.48 | 0.83 | 0.1 | |
| ARBackbone=Qwen3-14B, Layers (L)=402026.02 | 22.45 | 1 | — | |
| SWIFTBackbone=Qwen3-14B, Layers (L)=402026.02 | 22.38 | 1.14 | 84.4 | |
| CLaSpBackbone=Qwen3-8B, Layers (L)=362026.02 | 21.84 | 1.14 | 92.4 | |
| CLaSpBackbone=Qwen3-32B, Layers (L)=642026.02 | 21.65 | 1.3 | 96.1 | |
| CLaSpBackbone=Qwen3-14B, Layers (L)=402026.02 | 21.14 | 1.19 | 92.3 | |
| SWIFTBackbone=Qwen3-8B, Layers (L)=362026.02 | 20.6 | 1.12 | 75 | |
| SWIFTBackbone=Qwen3-32B, Layers (L)=642026.02 | 20.38 | 1.23 | 84.6 | |
| DELBackbone=Qwen3-32B, Layers (L)=642026.02 | 19.69 | 0.85 | 0 | |
| ARBackbone=Qwen3-32B, Layers (L)=642026.02 | 19.65 | 1 | — |