<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>LoRA on PlumePHP</title><link>https://plumephp.com/tags/lora/</link><description>Recent content in LoRA on PlumePHP</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Sun, 27 Sep 2026 11:30:00 +0800</lastBuildDate><atom:link href="https://plumephp.com/tags/lora/index.xml" rel="self" type="application/rss+xml"/><item><title>大模型高效微调：从全参微调到 LoRA/QLoRA 与 DPO</title><link>https://plumephp.com/ai-llm-fine-tuning/</link><pubDate>Sun, 27 Sep 2026 10:30:00 +0800</pubDate><guid>https://plumephp.com/ai-llm-fine-tuning/</guid><description>大模型高效微调：LoRA/QLoRA 原理、PEFT 技术、微调数据构造、SFT 与 DPO 对齐流程与工程实践。</description></item><item><title>量化感知训练（QAT）与量化微调：伪量化、STE 与 QLoRA 实战</title><link>https://plumephp.com/ai-qat-quantization-aware/</link><pubDate>Sun, 27 Sep 2026 11:30:00 +0800</pubDate><guid>https://plumephp.com/ai-qat-quantization-aware/</guid><description>&lt;blockquote&gt;
&lt;p&gt;训练后量化（PTQ）在精度敏感场景常常失手，量化感知训练（QAT）则让模型在训练阶段就&amp;quot;学会&amp;quot;适应量化噪声，是工业界把 INT8 精度损失压到 1% 以内的关键手段。本文从伪量化算子与直通估计器（STE）讲起，给出 QAT 完整流程，讲解蒸馏+量化组合，并深入 LoRA 量化微调（QLoRA）与工业实践。&lt;/p&gt;</description></item></channel></rss>