Needle-in-a-Haystack retrieval on NIAH-Single-3 128K context
98.8AccuracyQwen3+YaRN
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3+YaRNBackbone=8B, LS. Tokens=-2026.06 | 98.8 | |
| FlashMorphBackbone=1.7B, LS. Tokens=20M2026.06 | 94.4 | |
| FlashMorphBackbone=8B, LS. Tokens=20M2026.06 | 92.4 | |
| HALOBackbone=8B, LS. Tokens=234M2026.06 | 75 | |
| HALOBackbone=1.7B, LS. Tokens=234M2026.06 | 67.4 | |
| HALOBackbone=0.6B, LS. Tokens=234M2026.06 | 57.2 | |
| PostNAS*Backbone=1.7B, LS. Tokens=50B2026.06 | 57.2 | |
| FlashMorphBackbone=0.6B, LS. Tokens=20M2026.06 | 45.6 | |
| UniformBackbone=8B, LS. Tokens=N/A2026.06 | 40.6 | |
| UniformBackbone=0.6B, LS. Tokens=N/A2026.06 | 28.8 | |
| Qwen3+YaRNBackbone=30B-A3B, LS. Tokens=-2026.06 | 26.6 | |
| FlashMorphBackbone=30B-A3B, LS. Tokens=20M2026.06 | 24.6 | |
| UniformBackbone=30B-A3B, LS. Tokens=N/A2026.06 | 18.4 | |
| UniformBackbone=1.7B, LS. Tokens=N/A2026.06 | 16.4 | |
| HALOBackbone=30B-A3B, LS. Tokens=234M2026.06 | 12 | |
| KL-LSBackbone=1.7B, LS. Tokens=20B2026.06 | 9.4 | |
| KL-LSBackbone=0.6B, LS. Tokens=20B2026.06 | 8.6 | |
| Qwen3+YaRNBackbone=1.7B, LS. Tokens=-2026.06 | 0.8 | |
| Qwen3Backbone=0.6B, LS. Tokens=-2026.06 | 0 | |
| Qwen3+YaRNBackbone=0.6B, LS. Tokens=-2026.06 | 0 | |
| Qwen3Backbone=1.7B, LS. Tokens=-2026.06 | 0 | |
| Qwen3Backbone=8B, LS. Tokens=-2026.06 | 0 | |
| Qwen3Backbone=30B-A3B, LS. Tokens=-2026.06 | 0 |