<?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[RAG 里 top_k 应该设多少？]]></title><description><![CDATA[<p dir="auto">top_k 设 3 回答缺证据，设 20 又很慢，还会混乱。大家怎么定？</p>
]]></description><link>https://localaihub.com/topic/72/rag-里-top_k-应该设多少</link><generator>RSS for Node</generator><lastBuildDate>Wed, 03 Jun 2026 18:50:48 GMT</lastBuildDate><atom:link href="https://localaihub.com/topic/72.rss" rel="self" type="application/rss+xml"/><pubDate>Mon, 04 May 2026 03:01:00 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Tue, 05 May 2026 00:22:00 GMT]]></title><description><![CDATA[<p dir="auto">top_k 是旋钮，不是质量保证。</p>
]]></description><link>https://localaihub.com/post/363</link><guid isPermaLink="true">https://localaihub.com/post/363</guid><dc:creator><![CDATA[阿航]]></dc:creator><pubDate>Tue, 05 May 2026 00:22:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 22:15:00 GMT]]></title><description><![CDATA[<p dir="auto">最后看 context precision。top_k 大但有效证据比例低，就是噪声。</p>
]]></description><link>https://localaihub.com/post/362</link><guid isPermaLink="true">https://localaihub.com/post/362</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Mon, 04 May 2026 22:15:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 19:31:00 GMT]]></title><description><![CDATA[<p dir="auto">可以。再加低置信兜底：证据不足就追问或拒答。</p>
]]></description><link>https://localaihub.com/post/361</link><guid isPermaLink="true">https://localaihub.com/post/361</guid><dc:creator><![CDATA[nora]]></dc:creator><pubDate>Mon, 04 May 2026 19:31:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 17:43:00 GMT]]></title><description><![CDATA[<p dir="auto">我准备按问题类型设置：事实查询少一点，综合解释多一点。</p>
]]></description><link>https://localaihub.com/post/360</link><guid isPermaLink="true">https://localaihub.com/post/360</guid><dc:creator><![CDATA[小李不困]]></dc:creator><pubDate>Mon, 04 May 2026 17:43:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 17:03:00 GMT]]></title><description><![CDATA[<p dir="auto">但展示引用不能漏关键事实。内部用了某块，结果不展示，也会被质疑。</p>
]]></description><link>https://localaihub.com/post/359</link><guid isPermaLink="true">https://localaihub.com/post/359</guid><dc:creator><![CDATA[rootless]]></dc:creator><pubDate>Mon, 04 May 2026 17:03:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 16:53:00 GMT]]></title><description><![CDATA[<p dir="auto">可以。给模型用于理解，展示给用户的是关键证据。别把中间过程全丢出来。</p>
]]></description><link>https://localaihub.com/post/358</link><guid isPermaLink="true">https://localaihub.com/post/358</guid><dc:creator><![CDATA[阿白]]></dc:creator><pubDate>Mon, 04 May 2026 16:53:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 16:08:00 GMT]]></title><description><![CDATA[<p dir="auto">引用和给模型的上下文可以不一样？</p>
]]></description><link>https://localaihub.com/post/357</link><guid isPermaLink="true">https://localaihub.com/post/357</guid><dc:creator><![CDATA[小树]]></dc:creator><pubDate>Mon, 04 May 2026 16:08:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 13:17:00 GMT]]></title><description><![CDATA[<p dir="auto">我们产品限制最多 4 个引用，但内部给模型 6 个 chunk。</p>
]]></description><link>https://localaihub.com/post/356</link><guid isPermaLink="true">https://localaihub.com/post/356</guid><dc:creator><![CDATA[米饭]]></dc:creator><pubDate>Mon, 04 May 2026 13:17:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 10:39:00 GMT]]></title><description><![CDATA[<p dir="auto">还有引用数量。答案引用 12 个来源，用户通常看不下去。</p>
]]></description><link>https://localaihub.com/post/355</link><guid isPermaLink="true">https://localaihub.com/post/355</guid><dc:creator><![CDATA[小路灯]]></dc:creator><pubDate>Mon, 04 May 2026 10:39:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 09:16:00 GMT]]></title><description><![CDATA[<p dir="auto">那先别改 top_k，先看切块。小块需要 parent chunk 或上下文扩展。</p>
]]></description><link>https://localaihub.com/post/354</link><guid isPermaLink="true">https://localaihub.com/post/354</guid><dc:creator><![CDATA[MingK]]></dc:creator><pubDate>Mon, 04 May 2026 09:16:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 08:59:00 GMT]]></title><description><![CDATA[<p dir="auto">我们 chunk 很小，top_5 经常拼不出完整答案。</p>
]]></description><link>https://localaihub.com/post/353</link><guid isPermaLink="true">https://localaihub.com/post/353</guid><dc:creator><![CDATA[小周]]></dc:creator><pubDate>Mon, 04 May 2026 08:59:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 08:17:00 GMT]]></title><description><![CDATA[<p dir="auto">top_k 没有固定值，取决于 chunk 大小、问题类型、上下文预算。</p>
]]></description><link>https://localaihub.com/post/352</link><guid isPermaLink="true">https://localaihub.com/post/352</guid><dc:creator><![CDATA[林小北]]></dc:creator><pubDate>Mon, 04 May 2026 08:17:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 06:32:00 GMT]]></title><description><![CDATA[<p dir="auto">没 reranker 时 top_k 大了确实会污染上下文。模型会在弱相关块里找理由。</p>
]]></description><link>https://localaihub.com/post/351</link><guid isPermaLink="true">https://localaihub.com/post/351</guid><dc:creator><![CDATA[nora]]></dc:creator><pubDate>Mon, 04 May 2026 06:32:00 GMT</pubDate></item><item><title><![CDATA[Reply to RAG 里 top_k 应该设多少？ on Mon, 04 May 2026 06:06:00 GMT]]></title><description><![CDATA[<p dir="auto">分阶段。向量召回 top_30，rerank 后给模型 5-8 个，不是直接把 30 个都塞进去。</p>
]]></description><link>https://localaihub.com/post/350</link><guid isPermaLink="true">https://localaihub.com/post/350</guid><dc:creator><![CDATA[阿航]]></dc:creator><pubDate>Mon, 04 May 2026 06:06:00 GMT</pubDate></item></channel></rss>