Unverified50% confidenceBenchmarkExact time
CriticGen生成的评分表相关性和覆盖度分别从3.33/4.03提升到3.97/4.24
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified评测设置了语义正确率、三次全对率和格式正确率三个维度指标62% similarUnverified引导模式下平均端到端准确率从28.6%提升到33.7%,歧义检测F1值从45.3%提升到64.9%62% similarUnverifiedCriticGen在基于标准的理由和可执行建议上的F1分数分别从0.6369/0.5994提升到0.7554/0.790062% similarUnverifiedLFM2.5在BFCL-v3和BFCL-v4工具调用评测中分别比上代提升了19和23个点60% similarUnverified投机解码在候选token接受率较高时每次前向传播有效输出token可达2~4个,整体生成速度提升1.5x~3x60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/893107API
curl https://kongchang.com/api/v1/knowledge/claims/893107MCP
get_claim(id=893107)