intel/auto-round：README 來源編輯指南

intel/auto-round 的 README 將專案描述為「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.」。本文只整理倉庫可直接核對的內容，不把 star、Fork 或宣傳語當成品質證明。README 在「🚀 What is AutoRound?」下寫到：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。這說明的是專案邊界，不是已完成的生產驗證。

從 README 的「🆕 What's New」與相關條目，可以先判斷它是否處理你的實際問題：[2026/06] AutoScheme has been refined to improve accuracy for gguf format. See AutoScheme Accuracy for details. This enhancement incurs additional tuning cost.。若需求不同，不應只因專案熱度就採用。本文保留原始專案名、命令與元件名，方便回到一手來源核對。 README 另外列出一項可核對的資訊：[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.。這類原文條目可用來設計試跑步驟，但不能取代實際環境測試。

README 將運作方式分散在「✨ Key Features」等段落。可確認的線索包括：✅ Ecosystem Integration directly works with Transformers, vLLM, SGLang and more.。本文不把未寫出的架構、效能或安全邊界補成結論；真正的執行鏈仍要配合目錄、設定檔與版本標籤檢查。

第一次安裝應從 README 指出的入口開始。目前可核對的命令是： # 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 如果倉庫沒有命令，本文不會自行編造步驟，而是建議先閱讀「🆕 What's New」，確認系統依賴、預設埠與首次初始化。

日常使用取決於專案文件。README 的「✨ Key Features」段落提到：✅ Multiple Formats Export Support AutoRound, AutoAWQ, AutoGPTQ, and GGUF for maximum compatibility. Details are shown in export formats。設定檔、環境變數、權限與資料目錄只在來源明確時才會記錄；沒有寫出的預設值，應在測試環境驗證並保留回滾副本。 同一部分也提到：[2026/06] AutoRound is now fully integrated into vLLM-Omni: vLLM blog。