Unverified50% confidenceBenchmarkExact time
在LongMemEval基准测试中，20个chunk时BM25与嵌入方案收敛于约95%准确率
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
Verified如果每步准确率为95%，经过20步循环后整体准确率会降至约36%74% similarUnverified在CLINC150基准测试上，80条微调数据可使向量路由达到85.9%的准确率，完整fine-tune可达96%以上72% similarUnverified4比特量化相较于FP16通常会导致0.5-2%的基准测试精度下降70% similarUnverified在同一复现实验中，ResNet-20测试准确率为90.4%，ResNet-56测试准确率为91.7%，训练准确率分别为97.4%和99.0%69% similarUnverified训练中保留集 FID 从 33.7 降到 27.0，FD-DINOv2 从 570 降到 218，基于检测器的物体准确率从 65% 提升到 90%69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/920797API
curl https://kongchang.com/api/v1/knowledge/claims/920797MCP
get_claim(id=920797)