Long-context evaluation on RULER 64k Zero-shot transfer
32.9AccuracyStill
Evaluation results
| Method | Links | |
|---|---|---|
| StillK (Cache budget)=1024, Number of seeds=32026.06 | 32.9 | |
| StillK (Cache budget)=2048, Number of seeds=32026.06 | 31.8 | |
| StillK (Cache budget)=328, Number of seeds=32026.06 | 31.7 | |
| StillK (Cache budget)=512, Number of seeds=32026.06 | 27.5 | |
| StillK (Cache budget)=164, Number of seeds=32026.06 | 27.5 | |
| KV-DistillK (Cache budget)=512, Number of seeds=32026.06 | 21.8 | |
| KV-DistillK (Cache budget)=1024, Number of seeds=32026.06 | 21 | |
| StillK (Cache budget)=82, Number of seeds=32026.06 | 20.6 | |
| KV-DistillK (Cache budget)=2048, Number of seeds=32026.06 | 20.2 | |
| KV-DistillK (Cache budget)=82, Number of seeds=32026.06 | 19.4 | |
| KV-DistillK (Cache budget)=328, Number of seeds=32026.06 | 19.3 | |
| KV-DistillK (Cache budget)=164, Number of seeds=32026.06 | 18.6 |