Long-Context Retrieval on RULER 64K
97.33ScoreHiLS-Attn HoPE-Q-Cal
Evaluation Results
| Method | Links | |
|---|---|---|
| HiLS-Attn HoPE-Q-CalTuning Protocol=Full param. CPT, Positional Encoding=HoPE-Q-Cal2026.07 | 97.33 | |
| Nemotron-3-Ultra 550B-A55B-Baseshots=0-shot2026.06 | 95.3 | |
| Kimi-K2 Baseshots=0-shot2026.06 | 93.79 | |
| DeepSeek-V3.2 Exp-Baseshots=0-shot2026.06 | 93.3 | |
| Mistral-Large-3 675B-Base-2512shots=0-shot2026.06 | 90.11 | |
| Hybrid-KSAsetting=Train-from-scratch2026.04 | 69.95 | |
| Hybrid-GDNsetting=Train-from-scratch2026.04 | 69.39 | |
| Fullsetting=Train-from-scratch2026.04 | 65.32 | |
| Hybrid-SCAsetting=Train-from-scratch2026.04 | 63.54 | |
| Hybrid-SWAsetting=Train-from-scratch2026.04 | 63.03 | |
| KSAsetting=Train-from-scratch2026.04 | 57.13 | |
| GLM-4.5 Baseshots=0-shot2026.06 | 16.12 | |
| HiLS-Attn RoPE-Q-CalTuning Protocol=Full param. CPT, Positional Encoding=RoPE-Q-Cal2026.07 | 4.33 | |
| Olmo3 512swaTuning Protocol=Full param. CPT2026.07 | 1.33 | |
| LMK token tuningTuning Protocol=Freezing param.2026.07 | 1 | |
| HiLS-Attn NoPE-Q-CalTuning Protocol=Full param. CPT, Positional Encoding=NoPE-Q-Cal2026.07 | 0.33 | |
| Olmo3 BaseTuning Protocol=Freezing param.2026.07 | 0 |