Unverified50% confidenceFactExact time
衡量嵌入空间各向异性的指标包括平均余弦相似度（理想各向同性空间中应趋近于0）、有效维度/谱分析，以及IsoScore（给出[0,1]区间得分）
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
分散损失：破解小语言模型嵌入坍缩的关键技术
hackernewshackernews7/3/2026
Related Claims
Unverified一致性比率的衡量方法包括精确一致率、相邻一致率（允许±1分偏差）、Cohen's Kappa 系数，连续分数还可使用 Pearson 相关系数或 Spearman 等级相关58% similarVerifiedNumerical gradient checking approximates gradients using the central difference formula (f(θ+ε) - f(θ-ε)) / (2ε) and compares them with analytical gradients; a relative error below 1e-5 is considered correct, while an error above 1e-3 almost certainly indicates a bug.56% similarUnverifiedEarly testing on the Opus model showed an exact string recognition rate of 0 out of 15, whereas after the release of Fable 5 the same test yielded 13 out of 15.55% similarVerifiedKV Cache大小约等于 2 × 层数 × 注意力头数 × 头维度 × 序列长度 × 精度字节数55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/259448API
curl https://kongchang.com/api/v1/knowledge/claims/259448MCP
get_claim(id=259448)