<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[LoRA 微调实战：用 $10 训练专属模型]]></title><description><![CDATA[<blockquote>
<p dir="auto">来源：AI 订阅指南</p>
</blockquote>
<p dir="auto">LoRA（Low-Rank Adaptation）让普通开发者也能微调大模型。</p>
<p dir="auto"><strong>为什么选择 LoRA？</strong></p>
<ul>
<li>全量微调 7B 模型需要 8×A100（每小时 $24）</li>
<li>LoRA 微调只需 1×RTX 3090（$0.5/小时）</li>
<li>效果接近全量微调的 90%</li>
</ul>
<p dir="auto"><strong>工具选择：</strong></p>
<ul>
<li><strong>LLaMA-Factory</strong>：图形界面，最易上手</li>
<li><strong>Axolotl</strong>：配置灵活，社区活跃</li>
<li><strong>PEFT</strong>：HuggingFace 官方库</li>
</ul>
<p dir="auto"><strong>数据准备</strong>：500-5000 条高质量样本即可。格式为 JSONL：</p>
<pre><code class="language-json">{"instruction": "...", "input": "...", "output": "..."}
</code></pre>
<hr />
<p dir="auto"><em>更多本地部署教程请关注 AI 订阅指南。</em></p>
]]></description><link>https://aspxai.com/topic/209/lora-微调实战-用-10-训练专属模型</link><generator>RSS for Node</generator><lastBuildDate>Mon, 22 Jun 2026 07:36:17 GMT</lastBuildDate><atom:link href="https://aspxai.com/topic/209.rss" rel="self" type="application/rss+xml"/><pubDate>Mon, 22 Jun 2026 02:58:37 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to LoRA 微调实战：用 $10 训练专属模型 on Mon, 22 Jun 2026 03:03:44 GMT]]></title><description><![CDATA[<p dir="auto">我们用过 pgvector + LangChain，效果不错但查询延迟有点高。</p>
]]></description><link>https://aspxai.com/post/1206</link><guid isPermaLink="true">https://aspxai.com/post/1206</guid><dc:creator><![CDATA[y]]></dc:creator><pubDate>Mon, 22 Jun 2026 03:03:44 GMT</pubDate></item><item><title><![CDATA[Reply to LoRA 微调实战：用 $10 训练专属模型 on Mon, 22 Jun 2026 03:03:43 GMT]]></title><description><![CDATA[<p dir="auto">ChromaDB 在小数据量下够用，数据量大了建议换 Milvus。</p>
]]></description><link>https://aspxai.com/post/1205</link><guid isPermaLink="true">https://aspxai.com/post/1205</guid><dc:creator><![CDATA[回眸彼岸]]></dc:creator><pubDate>Mon, 22 Jun 2026 03:03:43 GMT</pubDate></item><item><title><![CDATA[Reply to LoRA 微调实战：用 $10 训练专属模型 on Mon, 22 Jun 2026 03:03:43 GMT]]></title><description><![CDATA[<p dir="auto">知识库更新频率也是个问题，我们做了增量索引方案。</p>
]]></description><link>https://aspxai.com/post/1204</link><guid isPermaLink="true">https://aspxai.com/post/1204</guid><dc:creator><![CDATA[画丹青海角]]></dc:creator><pubDate>Mon, 22 Jun 2026 03:03:43 GMT</pubDate></item><item><title><![CDATA[Reply to LoRA 微调实战：用 $10 训练专属模型 on Mon, 22 Jun 2026 03:03:43 GMT]]></title><description><![CDATA[<p dir="auto">ChromaDB 在小数据量下够用，数据量大了建议换 Milvus。</p>
]]></description><link>https://aspxai.com/post/1203</link><guid isPermaLink="true">https://aspxai.com/post/1203</guid><dc:creator><![CDATA[安年]]></dc:creator><pubDate>Mon, 22 Jun 2026 03:03:43 GMT</pubDate></item><item><title><![CDATA[Reply to LoRA 微调实战：用 $10 训练专属模型 on Mon, 22 Jun 2026 03:03:43 GMT]]></title><description><![CDATA[<p dir="auto">显存只有 8G 能跑什么模型？主要用于代码辅助。</p>
]]></description><link>https://aspxai.com/post/1202</link><guid isPermaLink="true">https://aspxai.com/post/1202</guid><dc:creator><![CDATA[品香茗酒肆]]></dc:creator><pubDate>Mon, 22 Jun 2026 03:03:43 GMT</pubDate></item></channel></rss>