AutoEMatic: Runtime-Adaptive Mixed-Precision LLM Inference via Automatic Exponent-Mantissa Refinement
arXiv:2611.10165v1 Announce Type: new Abstract: Existing LLM post-training quantization techniques determine precision before inference begins, yet the precision a model requires varies dynamically across prompts, layers, and inference phases. We present AutoEMatic, a full-stack co-design that adapts the precision of weights, activations, and the KV cache at runtime. A lightweight policy selects each layer's precision at every prefill and decode step from two runtime signals: the entropy of the
阅读原始来源
来源记录
研究分析
研究主张
论文标识：2611.10165。 来源描述：arXiv:2611.10165v1 Announce Type: new Abstract: Existing LLM post-training quantization techniques determine precision before inference begins, yet the precision a model requires varies dynamically across prompts, layers, and inference phases. We present AutoEMatic, a full-stack co-design that adapts the precision of weights, activations, and the KV cache at runtime. A lightweight policy selects each layer's precision at every prefill and decode step from two runtime signals: the entropy of the
评测与比较记录
可供比较的近期同类记录：“Conformal Fair Decoding for Large Language Model Recommenders” (arXiv, 2026年11月10日)；“The Infrastructure Divide: Analyzing Community Stance on Data Center Development via Reddit” (arXiv, 2026年11月10日)。主题相近仅用于导航，未核验其是否属于同一需求或独立证据。
结论适用范围
结论只能在所述模型版本、任务、数据与推理预算内解释。社区票数、下载量或榜单位置不能证明另一工作流中的可靠性。
有效性与复现缺口
需核对样本选择、基线配置、不确定性、数据污染风险、负面结果以及代码或权重可获取性；此处没有本站复现实验记录。
复核路线
阅读论文或模型卡及评测附录，针对明确基线复现一个有限结论，再测试代表性用户任务并记录结论失效的场景。