https://agi-talks.vercel.app/langchain-biweekly/assets/md-D7ldu_FY.js

1import{_ as c}from"./T-BFuaxOr9.js";import{o as r,c as i,k as a,l as _,a9 as t,q as p,s as f,A as s,e}from"./modules/vue-B6eO-bC7.js";import{_ as m}from"./slidev/iframe-right.vue_vue_type_script_setup_true_lang-DAnjngo4.js";import{u as d,f as h}from"./slidev/context-CCXkfbx0.js";import"./index-D1ZjaNgL.js";import"./modules/shiki-ChG2S6gQ.js";const u=e("h1",null,"🎙️ 少样本助力工具调用",-1),$=e("p",null,"实验表明 Few Shot 可显著提高工具调用准确性",-1),g=e("p",null,"我们在两个数据集上进行了实验：",-1),w=e("ul",null,[e("li",null,[e("p",null,[t("第一个，"),e("a",{href:"https://smith.langchain.com/public/6f62ae8b-4d96-4f0f-8eef-177ae3e30a65/d",target:"_blank"},"查询分析"),t(" 数据集，是一个相当标准的设置，其中使用对 LLM 的单次调用根据用户问题调用不同的搜索索引。")])]),e("li",null,[e("p",null,[t("第二个，"),e("a",{href:"https://smith.langchain.com/public/f8b159be-89e4-4f9f-93a9-30434fd31cbf/d",target:"_blank"},"多元宇宙数学"),t("，测试了在更具有 Agent 形态的 ReAct 工作流程的上下文中的函数调用（这涉及到对 LLM 的多次调用）。")])])],-1),x=e("p",null,"我们在多个 OpenAI 和 Anthropic 模型上进行了基准测试。我们尝试了不同的方法向模型提供少量示例，目的是看看哪种方法能产生最佳结果。",-1),b={__name:"slides.md__slidev_53",setup(k){const{$slidev:A,$nav:v,$clicksContext:o,$clicks:C,$page:L,$renderContext:y,$frontmatter:n}=d();return o.setup(),(B,N)=>{const l=c;return r(),i(m,p(f(s(h)(s(n),52))),{default:a(()=>[u,$,g,w,x,_(l,null,{default:a(()=>[t("We ran a few experiments, which show how few-shot prompting can significantly enhance model accuracy - especially for complex tasks. Read on for how we did it (and the results).")]),_:1})]),_:1},16)}}},q=b;export{q as default};