Mathematical Reasoning on GSM8K (Strict Accuracy, ALm, Rej)
93.1Accuracy (Strict)Cactus 1.0
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Cactus 1.0m (draft length)=10, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 93.1 | 5.44 | -32 | |
| TASm (draft length)=20, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 92.87 | 6.78 | -32 | |
| Cactus 1.0m (draft length)=20, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 92.87 | 7 | -34 | |
| TASm (draft length)=10, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 92.65 | 5.24 | -30 | |
| Cactus 0.75m (draft length)=20, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 92.15 | 7.15 | -36 | |
| Cactus 0.75m (draft length)=10, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 92.12 | 5.35 | -31 | |
| SpSm (draft length)=20, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 91.89 | 5.11 | — | |
| VerifierVerifier model=Qwen 3 14B2026.04 | 91.71 | — | — | |
| SpSm (draft length)=10, Verifier model=Qwen 3 14B, Drafter model=Qwen 3 0.6B2026.04 | 91.12 | 4.27 | — | |
| Cactus 0.75m (draft length)=20, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 86.66 | 7.5 | -37 | |
| TASm (draft length)=10, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 86.58 | 5.49 | -29 | |
| Cactus 1.0m (draft length)=20, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 86.43 | 7.61 | -39 | |
| Cactus 1.0m (draft length)=10, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 86.35 | 5.72 | -37 | |
| Cactus 0.75m (draft length)=10, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 85.97 | 5.65 | -34 | |
| TASm (draft length)=20, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 85.51 | 7.23 | -35 | |
| SpSm (draft length)=20, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 84.46 | 5.44 | — | |
| VerifierVerifier model=Qwen 3 8B2026.04 | 84.31 | — | — | |
| SpSm (draft length)=10, Verifier model=Qwen 3 8B, Drafter model=Qwen 3 0.6B2026.04 | 83.78 | 4.49 | — |