Mathematical Reasoning on AIME 2024 (Pass@1 with 1K, 2K, 4K Budgets)
54.5Pass@1 (1K Budget)ForesightKV
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ForesightKVModel=Qwen3-4B2026.02 | 54.5 | 70.2 | 69.2 | |
| ForesightKVModel=DeepSeek-R1-Distill-Qwen-7B2026.02 | 44.6 | 52.9 | 54.3 | |
| RPCModel=Qwen3-4B2026.02 | 10 | 36.7 | 59.2 | |
| DuoAttentionModel=DeepSeek-R1-Distill-Qwen-7B2026.02 | 3.3 | 13.3 | — |