1
Massive Spikes in LLMs are Bias Vectors: Mechanistic Uncovering and Spike-Free Quantization
颠覆认知：LLM中的巨大激活尖峰并非噪声，而是模型学习的偏差向量，并基于此提出无尖峰量化方法。
arXiv:2606.02288v1 Announce Type: new Abstract: Massive activation spikes in Large Language Models (LLMs) severely degrade quantization by stretching …