Code Generation on MBPP (Score, Relative Change, Avg. Gate Density)
39.2ScoreVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaAttention Mechanism=Full Attention, Model Size=8.1B, Training Context=32k2026.05 | 39.2 | — | — | |
| Self-Pruned KVAttention Mechanism=Self-Pruned KV, tau=0.5, Model Size=8.1B, Training Context=32k2026.05 | 37.2 | -5.1 | 24.8 |