Unverified50% confidenceBenchmarkExact time
论文测量的一次top-5发现交互仅使用475个token，而在全目录核算方式下同样操作需要约42,450个token
1
Sources
50%
Confidence
Long-term
Relevance
9/28/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在同一套简洁提示词和参考重叠评分器下，已发布的406M专用模型在ROUGE-1上至少领先五个商用托管大模型6.2分64% similarUnverified在MoE模型中每个token只激活少量Expert（通常2-4个，Top-K路由），实际计算量可能仅相当于6-7B的Dense模型61% similarUnverified最终结果为6/6数值测试通过，0个依赖漏洞，转换1,775个模块，生产JavaScript约213.6 kB，Gzip后约68.5 kB59% similarUnverifiedMD5生成的16位十六进制截取值提供约2^64种组合，被认为对软件授权场景的唯一性需求足够57% similarUnverifiedLFM2.5 has 8.3 billion total parameters but only 1.5 billion parameters activated per inference57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/955623API
curl https://kongchang.com/api/v1/knowledge/claims/955623MCP
get_claim(id=955623)