Mathematical Reasoning on MATH 500 (Accuracy, Throughput, Speedup)
69.6AccuracyFull Attn.
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full Attn.KV Budget=–, Backbone=Qwen3-8B2026.04 | 69.6 | 222.8 | — | |
| TriAttn.KV Budget=1024, Backbone=Qwen3-8B2026.04 | 68.4 | 1,405.2 | 6.3 |
| Method | Links | |||
|---|---|---|---|---|
| Full Attn.KV Budget=–, Backbone=Qwen3-8B2026.04 | 69.6 | 222.8 | — | |
| TriAttn.KV Budget=1024, Backbone=Qwen3-8B2026.04 | 68.4 | 1,405.2 | 6.3 |