Long-context retrieval on Needle-in-a-Haystack 1.0 (test)
99.9ScoreFastKV
Evaluation Results
| Method | Links | |
|---|---|---|
| FastKVModel=LLaMA-3.1-8B-Instruct, Prefill=60%, KV=10%2025.02 | 99.9 | |
| Full-contextModel=LLaMA-3.1-8B-Instruct, Prefill=100%, KV=100%2025.02 | 99 | |
| SnapKVModel=LLaMA-3.1-8B-Instruct, Prefill=100%, KV=10%2025.02 | 99 | |
| GemFilterModel=LLaMA-3.1-8B-Instruct, Prefill=51%, KV=10%2025.02 | 95.8 | |
| StreamingLLMModel=LLaMA-3.1-8B-Instruct, Prefill=100%, KV=10%2025.02 | 33.5 |