intel/auto-round：README 来源编辑指南

intel/auto-round 的 README 将项目描述为"A SOTA quantization algorithm for high-accuracy low-bit LLM inference, directly optimized for CPU/XPU/CUDA, with multi-datatype support and full compatibility with vLLM, SGLang, and Transformers."。本文只整理仓库能直接核验的内容，不把星标、Fork 或宣传语当成质量证明。README 在"🚀 What is AutoRound?"下的说明是：AutoRound is an advanced quantization toolkit designed for Large Language Models (LLMs) and Vision-Language Models (VLMs). It achieves high accuracy at ultra-low bit widths (2,4 bits) with minimal tuning by leveraging sign-gradient descent。这给出的首先是项目边界，而不是已经完成的生产验证。

从 README 的"🆕 What's New"和相关条目看，读者可以先判断它是否解决自己的具体问题：[2026/06] AutoScheme has been refined to improve accuracy for gguf format. See AutoScheme Accuracy for details. This enhancement incurs additional tuning cost.。如果你的目标与这段说明不一致，就不应仅凭项目热度采用它。这里保留原项目名、命令和组件名，方便回到一手来源核对。 README 还列出了另一条可核对的信息：[2026/07] torch.compile is enabled by default except on Windows to accelerate quantization. Minor numerical differences compared with the non-compiled path are expected due to compiler optimizations.。这类原文条目可以帮助读者设计试运行步骤，但不能代替自己的环境测试。

README 把工作方式分散写在"✨ Key Features"等段落中。可确认的线索包括：✅ Ecosystem Integration directly works with Transformers, vLLM, SGLang and more.。这篇整理没有把未写出的架构、性能或安全边界补成结论；真正的运行链仍应结合仓库目录、配置文件和版本标签检查。

第一次安装应从 README 给出的入口开始。当前可复核的命令是： # CPU(Xeon)/GPU(CUDA) pip install auto-round # CPU(Xeon)/GPU(CUDA) nightly pip install auto-round-nightly # HPU(Gaudi) # install inside the hpu docker container, e.g. vault.habana.ai/gaudi-docker/1.23.0/ubuntu24.04/habanalabs/pytorch-installer-2.9.0:latest pip install auto-round-hpu # XPU(Intel GPU) pip install torch --index-url https://download.pytorch.org/whl/xpu pip install auto-round 如果仓库没有提供命令，本文不会替它编造安装步骤，而是建议先打开 README 的"🆕 What's New"部分，确认系统依赖、默认端口和首次初始化动作。