AI Research · research
Detecting a Route Flip Is Easier Than Knowing Whether to Fix It: Causal Route-Mediated Damage in Qua...
论文摘要:Top-k Mixture-of-Experts (MoE) routing is discontinuous, so a deployment-motivated numerical disturbance -- simulated 4-bit KV-cache quantization read by a protected BF16 gate...
为什么重要
该论文来自一手研究源并通过 3qk7 Research Radar 的新鲜度、来源与研究价值筛选;后续代码、模型权重、复现和相关产品事件可继续关联到该记录。
检测到的变化
- 研究论文进入 3qk7 高价值 Research Radar
一手证据
- arXiv cs.AI · 2026-08-14T03:53:00Z