<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>Ai on 编程心语</title>
		<link>https://www.ithome.me/tags/ai/</link>
		<description>Recent content in Ai on 编程心语</description>
		<generator>Hugo</generator>
		<language>zh-CN</language>
		
		
		
		
			<lastBuildDate>Fri, 07 Aug 2026 12:30:47 +0800</lastBuildDate>
		
			<atom:link href="https://www.ithome.me/tags/ai/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>端侧 Embedding 模型实战：为本地 RAG 选对编码器</title>
				<link>https://www.ithome.me/post/2026/08/07/edge-embedding-models-comparison/</link>
				<pubDate>Fri, 07 Aug 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/08/07/edge-embedding-models-comparison/</guid>
				<description>&lt;h2 id=&#34;为什么-embedding-模型是-rag-的地基&#34;&gt;为什么 Embedding 模型是 RAG 的&amp;quot;地基&amp;quot;&lt;/h2&gt;&#xA;&lt;p&gt;做本地 RAG（检索增强生成）时，大部分人把精力花在 LLM 选型上——7B 还是 13B？量化到 4bit 够不够？但真正决定 RAG 检索质量的，往往是管道入口处的 Embedding 模型。&lt;/p&gt;</description>
			</item>
			<item>
				<title>MCP 实战：把你的本地服务变成 AI Agent 的工具</title>
				<link>https://www.ithome.me/post/2026/08/06/mcp-practical-guide/</link>
				<pubDate>Thu, 06 Aug 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/08/06/mcp-practical-guide/</guid>
				<description>&lt;h2 id=&#34;为什么-agent-需要-mcp&#34;&gt;为什么 Agent 需要 MCP？&lt;/h2&gt;&#xA;&lt;p&gt;想象你是老板，想雇一个全能的 AI 助手帮你管服务器。问题来了：助手很聪明，但&lt;strong&gt;不会开服务器&lt;/strong&gt;——它看不到你的服务，也没法操作。&lt;/p&gt;&#xA;&lt;p&gt;MCP（Model Context Protocol）就是&lt;strong&gt;给 AI 装上手&lt;/strong&gt;的标准协议：把本地服务的 API 包装成 AI 能看懂、能调用的工具列表。2025 年 Anthropic 开源后，已经成为 Agent 生态的事实标准。&lt;/p&gt;</description>
			</item>
			<item>
				<title>LLM API 缓存实战：三层缓存把大模型调用成本打下来</title>
				<link>https://www.ithome.me/post/2026/08/04/llm-api-cache-practical-guide/</link>
				<pubDate>Tue, 04 Aug 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/08/04/llm-api-cache-practical-guide/</guid>
				<description>&lt;h2 id=&#34;为什么你的-llm-账单越烧越高&#34;&gt;为什么你的 LLM 账单越烧越高&lt;/h2&gt;&#xA;&lt;p&gt;接入大模型三个月，账单比想象中高出一大截。翻日志发现一个扎心的事实：&lt;strong&gt;大量请求是重复的&lt;/strong&gt;。&lt;/p&gt;&#xA;&lt;ul&gt;&#xA;&lt;li&gt;定时任务每天早上用同一份模板汇总数据，prompt 一字不差，却每次全价调用；&lt;/li&gt;&#xA;&lt;li&gt;用户提问&amp;quot;帮我解释下这段代码&amp;quot;和&amp;quot;这段代码什么意思&amp;quot;，语义相同，模型却要重新推理一遍；&lt;/li&gt;&#xA;&lt;li&gt;长 system prompt（几千 token 的角色设定 + 知识库）每次请求都重复计费。&lt;/li&gt;&#xA;&lt;/ul&gt;&#xA;&lt;p&gt;LLM 按 token 计费，重复推理就是纯浪费。本文分享一套实战验证过的三层缓存方案，从上到下依次是：&lt;strong&gt;精确缓存 → 语义缓存 → 供应商 Prompt Cache&lt;/strong&gt;，三层叠加通常能把成本降 50% 以上，顺带把响应延迟从秒级降到毫秒级。&lt;/p&gt;</description>
			</item>
			<item>
				<title>LLM 结构化输出实战：用 JSON Schema 约束大模型返回可靠数据</title>
				<link>https://www.ithome.me/post/2026/08/02/llm-structured-output-json-schema/</link>
				<pubDate>Sun, 02 Aug 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/08/02/llm-structured-output-json-schema/</guid>
				<description>&lt;h2 id=&#34;为什么需要结构化输出&#34;&gt;为什么需要结构化输出&lt;/h2&gt;&#xA;&lt;p&gt;接入大模型后最头疼的问题：你让它&amp;quot;提取订单信息&amp;quot;，它给你返回一段带语气词、Markdown 代码块、甚至把键名从 &lt;code&gt;order_id&lt;/code&gt; 改成 &lt;code&gt;orderId&lt;/code&gt; 的自由文本。下游程序一解析就崩。&lt;/p&gt;</description>
			</item>
			<item>
				<title>用 Ollama &#43; Continue 搭建本地代码审查助手</title>
				<link>https://www.ithome.me/post/2026/07/26/local-code-review-ollama-continue/</link>
				<pubDate>Sun, 26 Jul 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/07/26/local-code-review-ollama-continue/</guid>
				<description>&lt;h2 id=&#34;为什么需要本地代码审查&#34;&gt;为什么需要本地代码审查&lt;/h2&gt;&#xA;&lt;p&gt;传统代码审查（Code Review）依赖人工逐行检查，耗时久、标准不统一。而托管的 AI 审查服务（如 GitHub Copilot、CodeRabbit）虽然强大，却存在两个痛点：&lt;/p&gt;</description>
			</item>
			<item>
				<title>从勘智云平台拆解：如何自建一套端到端AI模型训练与部署平台</title>
				<link>https://www.ithome.me/post/2026/07/15/edge-ai-training-platform-architecture/</link>
				<pubDate>Wed, 15 Jul 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/07/15/edge-ai-training-platform-architecture/</guid>
				<description>剖析勘智云平台的七种CV任务、数据标注、模型训练、KModel部署全流程，给出基于开源工具链的自建方案。</description>
			</item>
			<item>
				<title>AI时代编程学习的思考：当语法不再稀缺，什么才值钱？</title>
				<link>https://www.ithome.me/post/2026/07/13/ai-era-programming-learning-reflections/</link>
				<pubDate>Mon, 13 Jul 2026 08:00:00 +0800</pubDate>
				<guid>https://www.ithome.me/post/2026/07/13/ai-era-programming-learning-reflections/</guid>
				<description>&lt;h2 id=&#34;当语法不再稀缺&#34;&gt;当语法不再稀缺&lt;/h2&gt;&#xA;&lt;p&gt;二十年前学编程，第一关是语法。C语言的指针、Java的泛型、Python的缩进——你要花大量时间记规则、查文档、对着报错信息发呆。那时候&amp;quot;会编程&amp;quot;的门槛，很大一部分是&amp;quot;记住了多少语法&amp;quot;。&lt;/p&gt;</description>
			</item>
	</channel>
</rss>
