In-Context Retrieval on RULER (test)
100S1 ScoreFull
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FullBackbone=Qwen3-8B, Context Window=32k2026.04 | 100 | 100 | 99.6 | 98.2 | 99.75 | 99.6 | 49 | 60.9 | 99.8 | 97.67 | 90.45 | |
| QuestBackbone=Qwen3-8B, Context Window=32k2026.04 | 100 | 30.2 | 28 | 4 | 11.25 | 5.4 | 32.6 | 39.35 | 50.48 | 87.87 | 38.92 | |
| DSBackbone=Qwen3-8B, Context Window=32k2026.04 | 100 | 100 | 99.6 | 98.2 | 99.65 | 99.75 | 49.2 | 62.92 | 99.76 | 95.27 | 90.44 | |
| AsyncTLSBackbone=Qwen3-8B, Context Window=32k2026.04 | 100 | 100 | 99.8 | 98.6 | 99.7 | 99.45 | 52.2 | 59.62 | 99.8 | 96.27 | 90.54 | |
| FullBackbone=Qwen3-14B, Context Window=32k2026.04 | 100 | 100 | 99.8 | 99.6 | 99.95 | 100 | 57.8 | 66.6 | 100 | 38.67 | 86.24 | |
| QuestBackbone=Qwen3-14B, Context Window=32k2026.04 | 100 | 34.6 | 41 | 5 | 17.85 | 10.2 | 35.6 | 44.37 | 40.24 | 50 | 37.89 | |
| DSBackbone=Qwen3-14B, Context Window=32k2026.04 | 100 | 34.6 | 100 | 99.8 | 99.95 | 99.85 | 55.6 | 65.7 | 99.92 | 16.6 | 77.2 | |
| AsyncTLSBackbone=Qwen3-14B, Context Window=32k2026.04 | 100 | 100 | 99.8 | 99.6 | 99.8 | 99.65 | 58.4 | 64.75 | 100 | 50.13 | 87.21 |