Long-context understanding on InfiniteBench v1 (test)
20DialogueSnapKV
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SnapKVBackbone=Llama-3.1-8B-UL, KV Budget=20482026.01 | 20 | — | — | — | — | 29 | 31 | 35 | 26 | 26 | 29 | 35 | 34 | 30 | 29 | |
| Llama-3.1-8BBase Model=Llama-3.1-8B, Tokens=128k2025.05 | 19 | 99.6 | 30.2 | 34 | 22.8 | 44 | — | — | — | — | — | — | — | — | — | |
| TailorKV-2Base Model=Llama-3.1-8B, Tokens=128(+896)2025.05 | 18.5 | 94.8 | 30 | 34 | 22.8 | 42.6 | — | — | — | — | — | — | — | — | — | |
| TailorKV-1Base Model=Llama-3.1-8B, Tokens=128(+896)2025.05 | 18 | 86.5 | 28.9 | 34 | 22.8 | 40.4 | — | — | — | — | — | — | — | — | — | |
| ASLBackbone=Llama-3.1-8B-UL, KV Budget=20482026.01 | 18 | — | — | — | — | 25 | 29 | 29 | 24 | 23 | 25 | 32 | 29 | 26 | 25 | |
| FastKVBackbone=Llama-3.1-8B-UL, KV Budget=20482026.01 | 17 | — | — | — | — | 25 | 27 | 31 | 23 | 23 | 25 | 32 | 29 | 27 | 25 | |
| GemFilterBackbone=Llama-3.1-8B-UL, KV Budget=20482026.01 | 17 | — | — | — | — | 23 | 32 | 30 | 21 | 23 | 24 | 25 | 22 | 25 | 21 | |
| ASL_2passBackbone=Llama-3.1-8B-UL, KV Budget=20482026.01 | 17 | — | — | — | — | 23 | 32 | 29 | 20 | 23 | 24 | 24 | 23 | 24 | 21 | |
| PQCacheBase Model=Llama-3.1-8B, Tokens=10242025.05 | 15 | 5.5 | 27.5 | 34 | 23.3 | 21.5 | — | — | — | — | — | — | — | — | — | |
| QuestBase Model=Llama-3.1-8B, Tokens=10242025.05 | 14 | 64.4 | 25.7 | 34 | 25.1 | 33.8 | — | — | — | — | — | — | — | — | — | |
| SnapKVBase Model=Llama-3.1-8B, Tokens=10242025.05 | 9.5 | 96.6 | 27.4 | 34 | 22.8 | 41 | — | — | — | — | — | — | — | — | — | |
| SnapKVBackbone=Qwen2.5-7B, KV Budget=20482026.01 | 9 | — | — | — | — | 13 | 16 | 12 | 12 | 12 | 13 | 18 | 15 | 15 | 15 | |
| FastKVBackbone=Qwen2.5-7B, KV Budget=20482026.01 | 9 | — | — | — | — | 13 | 16 | 12 | 12 | 12 | 13 | 18 | 15 | 15 | 15 | |
| ASLBackbone=Qwen2.5-7B, KV Budget=20482026.01 | 9 | — | — | — | — | 13 | 16 | 12 | 12 | 12 | 12 | 18 | 15 | 15 | 15 | |
| StreamLLMBase Model=Llama-3.1-8B, Tokens=10242025.05 | 7 | 3.2 | 23.7 | 34 | 22.8 | 18.3 | — | — | — | — | — | — | — | — | — | |
| GemFilterBackbone=Qwen2.5-7B, KV Budget=20482026.01 | 7 | — | — | — | — | 10 | 15 | 9 | 9 | 9 | 9 | 13 | 7 | 11 | 11 | |
| ASL_2passBackbone=Qwen2.5-7B, KV Budget=20482026.01 | 7 | — | — | — | — | 10 | 14 | 9 | 9 | 9 | 9 | 14 | 6 | 11 | 12 | |
| TailorKV-2Base Model=Yi-9B, Tokens=128(+896)2025.05 | 4.5 | 98.5 | 25.3 | 24 | 24.9 | 39.4 | — | — | — | — | — | — | — | — | — | |
| QuestBase Model=Yi-9B, Tokens=10242025.05 | 4 | 98.4 | 21.8 | 18.2 | 18.7 | 36.1 | — | — | — | — | — | — | — | — | — | |
| SnapKVBase Model=Yi-9B, Tokens=10242025.05 | 3 | 59 | 24.9 | 22.5 | 26.6 | 30 | — | — | — | — | — | — | — | — | — | |
| QuestBase Model=Yi-6B, Tokens=10242025.05 | 3 | 99.5 | 22 | 5.1 | 26.6 | 35.8 | — | — | — | — | — | — | — | — | — | |
| TailorKV-2Base Model=Yi-6B, Tokens=128(+896)2025.05 | 3 | 98.5 | 25.3 | 8 | 26.7 | 37.3 | — | — | — | — | — | — | — | — | — | |
| Yi-9BBase Model=Yi-9B, Tokens=200k2025.05 | 2.5 | 100 | 25.2 | 23.4 | 26.3 | 39.2 | — | — | — | — | — | — | — | — | — | |
| StreamLLMBase Model=Yi-9B, Tokens=10242025.05 | 2.5 | 1.5 | 24.2 | 23.7 | 21.3 | 16.4 | — | — | — | — | — | — | — | — | — | |
| TailorKV-1Base Model=Yi-9B, Tokens=128(+896)2025.05 | 2.5 | 98.7 | 26.6 | 24 | 21.3 | 39.2 | — | — | — | — | — | — | — | — | — | |
| SnapKVBase Model=Yi-6B, Tokens=10242025.05 | 2.5 | 55.7 | 23.2 | 4.5 | 26.9 | 26.4 | — | — | — | — | — | — | — | — | — | |
| TailorKV-1Base Model=Yi-6B, Tokens=128(+896)2025.05 | 2.5 | 98.7 | 25.3 | 7.7 | 26.4 | 37.2 | — | — | — | — | — | — | — | — | — | |
| PQCacheBase Model=Yi-9B, Tokens=10242025.05 | 2 | 7.8 | 25.3 | 22.2 | 25.6 | 18.5 | — | — | — | — | — | — | — | — | — | |
| PQCacheBase Model=Yi-6B, Tokens=10242025.05 | 1.5 | 6.9 | 24.6 | 5.7 | 26.9 | 16.3 | — | — | — | — | — | — | — | — | — | |
| Yi-6BBase Model=Yi-6B, Tokens=200k2025.05 | 0 | 99.2 | 25.2 | 6.8 | 26.9 | 37.1 | — | — | — | — | — | — | — | — | — | |
| StreamLLMBase Model=Yi-6B, Tokens=10242025.05 | 0 | 2 | 21.8 | 4.8 | 25.8 | 13.5 | — | — | — | — | — | — | — | — | — |