thexpin· @thexpin · X·· 8 小时前AI 评分
Seed 团队预印本报告 DeepSeek V4 存在 KV-cache 相位敏感性
Seed 团队一篇预印本报告 DeepSeek V4 和 V4.1-Flash 存在相位敏感性：相同信息在压缩 KV-cache 块中的位置不同，检索难度也不同。该技术可降低内存和注意力开销，但长上下文检索准确率在不同位置间差异最多达 40 个百分点；作者指出平均分会掩盖这类反复出现的弱点，且该发现针对检索问题，并非解释所有已报告的模型异常。论文见 arxiv.org/abs/2609.36322。
A Seed team preprint reports “phase sensitivity” in DeepSeek V4 and V4.1-Flash: identical information can become harder to retrieve depending on its position within compressed KV-cache blocks.
The technique cuts memory and attention costs, but long-context retrieval accuracy varied by up to 40 percentage points across positions.
Average benchmark scores can hide these recurring weak spots. The findings concern retrieval, not every reported model glitch.
arxiv.org/abs/2609.36322
来源：thexpin · x.com