Question Answering on RULER-HQA 112K context length
80.47Normalized Sub-EM AccuracyLycheeMemory-7B w/o Gate
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LycheeMemory-7B w/o GateModel=LycheeMemory-7B, Gate=disabled2026.02 | 80.47 | — | |
| RL-MemAgent-7BModel=RL-MemAgent-7B2026.02 | 79.69 | — | |
| LycheeMemory-7BModel=LycheeMemory-7B, Gate=enabled2026.02 | 75.78 | 3.5 | |
| Search-R1Model=Search-R12026.02 | 66.67 | — | |
| RAG + Qwen2.5-7B-InstructModel=RAG + Qwen2.5-7B-Instruct2026.02 | 64.84 | — | |
| Qwen2.5-Instruct-7B-1MModel=Qwen2.5-Instruct-7B-1M2026.02 | 51.56 | — | |
| Qwen2.5-Instruct-14B-1MModel=Qwen2.5-Instruct-14B-1M2026.02 | 50 | — | |
| QwenLong-L1-32BModel=QwenLong-L1-32B2026.02 | 31.25 | — | |
| DS-Distill-Qwen-32BModel=DS-Distill-Qwen-32B2026.02 | 23.44 | — | |
| DS-Distill-Qwen-14BModel=DS-Distill-Qwen-14B2026.02 | 14.84 | — | |
| DS-Distill-Qwen-7BModel=DS-Distill-Qwen-7B2026.02 | 0 | — |