Long-context retrieval and aggregation on RULER transfer mixed-domain (32k, 64k, 128k context)
37.2Performance Score (32k Context)Still
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| StillDecoding=deterministic greedy generation, max_new_tokens=192, repetition penalty=1.2, no-repeat 4-gram blocking=true2026.06 | 37.2 | 32.9 | 31.3 | |
| KV-DistillDecoding=deterministic greedy generation, max_new_tokens=192, repetition penalty=1.2, no-repeat 4-gram blocking=true2026.06 | 24.1 | 21.8 | 22 |