Long-context retrieval on NIAH (Δacc)
0Delta Accuracy (Δacc)Runtime-Certified Bounded-Error Quantized Attention
Evaluation Results
| Method | Links | |
|---|---|---|
| Runtime-Certified Bounded-Error Quantized AttentionContext Length=8K2026.05 | 0 | |
| Runtime-Certified Bounded-Error Quantized AttentionContext Length=64K2026.05 | 0 | |
| Runtime-Certified Bounded-Error Quantized AttentionContext Length=32K2026.05 | -2 |