Mathematical Reasoning on GSM8k (Score, Relative Change, Gate Density)
0.292ScoreVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaAttention Mechanism=Full Attention, Model Size=8.1B, Training Context=32k2026.05 | 0.292 | — | — | |
| Self-Pruned KVAttention Mechanism=Self-Pruned KV, tau=0.5, Model Size=8.1B, Training Context=32k2026.05 | 0.287 | -1.6 | 18.7 |