<?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[DSH 中启用自定义后端的思维深度（reasoning effort）]]></title><description><![CDATA[<h1>DSH 中启用自定义后端的思维深度（reasoning effort）</h1>
<hr />
<h2>UI效果展示</h2>
<p dir="auto"><img src="https://upload.lcz.me/uploads/f835efe2-982f-4ad0-ad35-e27b4b6c498c.jpeg" alt="7b37d1fd-d2b6-47b3-ad4c-6a18850eeb7a-image.jpeg" class=" img-fluid img-markdown" /></p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/c0677587-15e9-4889-9f07-ed3d56deedb5.jpeg" alt="5c860b40-ded7-476a-8de2-cc4080e3fc7e-image.jpeg" class=" img-fluid img-markdown" /></p>
<hr />
<h2>1. 配置步骤</h2>
<h3>1.1 修改配置文件</h3>
<p dir="auto"><code>/root/.dsh/settings.yaml</code>（或对应 profile 的 settings.yaml）：</p>
<pre><code class="language-yaml">llm-pi-ai:
  providers:
    sglang:
      displayName: sglang
      apiKeyEnv: SGLANG_API_KEY
      api: openai-completions
      baseURL: http://127.0.0.1:18080/v1
      models:
        - id: Qwen3.8-27B-AWQ
          name: Qwen3.8-27B-AWQ
          input: [ text, image ]
          contextWindow: 256000
          maxTokens: 32000
          # key = DSH 逻辑档位（off|minimal|low|medium|high|xhigh|max）
          # value = 发给后端的 wire 值；只有 off 允许留空
          reasoningEfforts:
            off: "none"      # 关思考：显式发 reasoning_effort="none"
            low: "low"
            medium: "medium"
            high: "xhigh"    # 注意：模板最高档叫 xhigh，不是 high
          # 该模型的后端模板不认 developer 角色时保留此项（与思考深度无关，纯兼容开关）
          compat:
            supportsDeveloperRole: false
</code></pre>
<h3>1.2 配置提供商默认挡位（可选）</h3>
<pre><code class="language-yaml">llm-pi-ai:
  providers:
    sglang:
      reasoning: medium     # 任一请求未显式带档位时，用 medium
</code></pre>
<h3>1.3 配置全局默认挡位</h3>
<pre><code class="language-yaml">agent-default-model:
  provider: sglang
  model: Qwen3.8-27B-AWQ
  reasoningEffort: medium   # 全局缺省，取值同逻辑档位集合
</code></pre>
<hr />
]]></description><link>https://lcz.me/topic/1179/dsh-中启用自定义后端的思维深度-reasoning-effort</link><generator>RSS for Node</generator><lastBuildDate>Sat, 22 Aug 2026 00:31:06 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/1179.rss" rel="self" type="application/rss+xml"/><pubDate>Tue, 18 Aug 2026 05:04:23 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to DSH 中启用自定义后端的思维深度（reasoning effort） on Wed, 19 Aug 2026 07:14:38 GMT]]></title><description><![CDATA[<p dir="auto">我大幅精简了内容，之前确实太臃肿了，感谢提醒。</p>
]]></description><link>https://lcz.me/post/12878</link><guid isPermaLink="true">https://lcz.me/post/12878</guid><dc:creator><![CDATA[neo]]></dc:creator><pubDate>Wed, 19 Aug 2026 07:14:38 GMT</pubDate></item><item><title><![CDATA[Reply to DSH 中启用自定义后端的思维深度（reasoning effort） on Tue, 18 Aug 2026 10:54:21 GMT]]></title><description><![CDATA[<p dir="auto">太长的文字，老哥你发点质量好点的截图，不然网站被谷歌判定为垃圾，我还想抢救下。</p>
]]></description><link>https://lcz.me/post/12730</link><guid isPermaLink="true">https://lcz.me/post/12730</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Tue, 18 Aug 2026 10:54:21 GMT</pubDate></item></channel></rss>