Spaces:

siro1
/

triton-for-llms

Running

App Files Files Community

triton-for-llms / src /_site /notebooks /rag_evaluation.html

siro1 HF Staff

commit from `quarto publish`

cf7a8a2 about 2 months ago

raw

history blame contribute delete

135 kB

	<!DOCTYPE html>
	<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>

	<meta charset="utf-8">
	<meta name="generator" content="quarto-1.6.40">

	<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">


	<title>RAG Evaluation – Open-Source AI Cookbook</title>
	<style>
	code{white-space: pre-wrap;}
	span.smallcaps{font-variant: small-caps;}
	div.columns{display: flex; gap: min(4vw, 1.5em);}
	div.column{flex: auto; overflow-x: auto;}
	div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
	ul.task-list{list-style: none;}
	ul.task-list li input[type="checkbox"] {
	width: 0.8em;
	margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */
	vertical-align: middle;
	}
	/* CSS for syntax highlighting */
	pre > code.sourceCode { white-space: pre; position: relative; }
	pre > code.sourceCode > span { line-height: 1.25; }
	pre > code.sourceCode > span:empty { height: 1.2em; }
	.sourceCode { overflow: visible; }
	code.sourceCode > span { color: inherit; text-decoration: inherit; }
	div.sourceCode { margin: 1em 0; }
	pre.sourceCode { margin: 0; }
	@media screen {
	div.sourceCode { overflow: auto; }
	}
	@media print {
	pre > code.sourceCode { white-space: pre-wrap; }
	pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
	}
	pre.numberSource code
	{ counter-reset: source-line 0; }
	pre.numberSource code > span
	{ position: relative; left: -4em; counter-increment: source-line; }
	pre.numberSource code > span > a:first-child::before
	{ content: counter(source-line);
	position: relative; left: -1em; text-align: right; vertical-align: baseline;
	border: none; display: inline-block;
	-webkit-touch-callout: none; -webkit-user-select: none;
	-khtml-user-select: none; -moz-user-select: none;
	-ms-user-select: none; user-select: none;
	padding: 0 4px; width: 4em;
	}
	pre.numberSource { margin-left: 3em; padding-left: 4px; }
	div.sourceCode
	{ }
	@media screen {
	pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
	}
	</style>


	<script src="../site_libs/quarto-nav/quarto-nav.js"></script>
	<script src="../site_libs/quarto-nav/headroom.min.js"></script>
	<script src="../site_libs/clipboard/clipboard.min.js"></script>
	<script src="../site_libs/quarto-search/autocomplete.umd.js"></script>
	<script src="../site_libs/quarto-search/fuse.min.js"></script>
	<script src="../site_libs/quarto-search/quarto-search.js"></script>
	<meta name="quarto:offset" content="../">
	<script src="../site_libs/quarto-html/quarto.js"></script>
	<script src="../site_libs/quarto-html/popper.min.js"></script>
	<script src="../site_libs/quarto-html/tippy.umd.min.js"></script>
	<script src="../site_libs/quarto-html/anchor.min.js"></script>
	<link href="../site_libs/quarto-html/tippy.css" rel="stylesheet">
	<link href="../site_libs/quarto-html/quarto-syntax-highlighting-549806ee2085284f45b00abea8c6df48.css" rel="stylesheet" id="quarto-text-highlighting-styles">
	<script src="../site_libs/bootstrap/bootstrap.min.js"></script>
	<link href="../site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
	<link href="../site_libs/bootstrap/bootstrap-2be10d9e998f81ff6e49e26833438aa5.min.css" rel="stylesheet" append-hash="true" id="quarto-bootstrap" data-mode="light">
	<script id="quarto-search-options" type="application/json">{
	"location": "sidebar",
	"copy-button": false,
	"collapse-after": 3,
	"panel-placement": "start",
	"type": "textbox",
	"limit": 50,
	"keyboard-shortcut": [
	"f",
	"/",
	"s"
	],
	"show-item-context": false,
	"language": {
	"search-no-results-text": "No results",
	"search-matching-documents-text": "matching documents",
	"search-copy-link-title": "Copy link to search",
	"search-hide-matches-text": "Hide additional matches",
	"search-more-match-text": "more match in this document",
	"search-more-matches-text": "more matches in this document",
	"search-clear-button-title": "Clear",
	"search-text-placeholder": "",
	"search-detached-cancel-button-title": "Cancel",
	"search-submit-button-title": "Submit",
	"search-label": "Search"
	}
	}</script>


	<link rel="stylesheet" href="../styles.css">
	</head>

	<body class="nav-sidebar docked">

	<div id="quarto-search-results"></div>
	<header id="quarto-header" class="headroom fixed-top">
	<nav class="quarto-secondary-nav">
	<div class="container-fluid d-flex">
	<button type="button" class="quarto-btn-toggle btn" data-bs-toggle="collapse" role="button" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
	<i class="bi bi-layout-text-sidebar-reverse"></i>
	</button>
	<nav class="quarto-page-breadcrumbs" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item">Open-Source AI Cookbook</li><li class="breadcrumb-item"><a href="../notebooks/rag_zephyr_langchain.html">RAG Techniques</a></li><li class="breadcrumb-item"><a href="../notebooks/rag_evaluation.html">RAG Evaluation</a></li></ol></nav>
	<a class="flex-grow-1" role="navigation" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item" aria-controls="quarto-sidebar" aria-expanded="false" aria-label="Toggle sidebar navigation" onclick="if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }">
	</a>
	<button type="button" class="btn quarto-search-button" aria-label="Search" onclick="window.quartoOpenSearch();">
	<i class="bi bi-search"></i>
	</button>
	</div>
	</nav>
	</header>
	<!-- content -->
	<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article">
	<!-- sidebar -->
	<nav id="quarto-sidebar" class="sidebar collapse collapse-horizontal quarto-sidebar-collapse-item sidebar-navigation docked overflow-auto">
	<div class="pt-lg-2 mt-2 text-left sidebar-header">
	<div class="sidebar-title mb-0 py-0">
	<a href="../">Open-Source AI Cookbook</a>
	</div>
	</div>
	<div class="mt-2 flex-shrink-0 align-items-center">
	<div class="sidebar-search">
	<div id="quarto-search" class="" title="Search"></div>
	</div>
	</div>
	<div class="sidebar-menu-container">
	<ul class="list-unstyled mt-1">
	<li class="sidebar-item sidebar-item-section">
	<div class="sidebar-item-container">
	<a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-1" role="navigation" aria-expanded="true">
	<span class="menu-text">About</span></a>
	<a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-1" role="navigation" aria-expanded="true" aria-label="Toggle section">
	<i class="bi bi-chevron-right ms-2"></i>
	</a>
	</div>
	<ul id="quarto-sidebar-section-1" class="collapse list-unstyled sidebar-section depth1 show">
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../index.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">About Quarto</span></a>
	</div>
	</li>
	</ul>
	</li>
	<li class="sidebar-item sidebar-item-section">
	<div class="sidebar-item-container">
	<a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true">
	<span class="menu-text">Open-Source AI Cookbook</span></a>
	<a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-2" role="navigation" aria-expanded="true" aria-label="Toggle section">
	<i class="bi bi-chevron-right ms-2"></i>
	</a>
	</div>
	<ul id="quarto-sidebar-section-2" class="collapse list-unstyled sidebar-section depth1 show">
	<li class="sidebar-item sidebar-item-section">
	<div class="sidebar-item-container">
	<a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true">
	<span class="menu-text">RAG Techniques</span></a>
	<a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-3" role="navigation" aria-expanded="true" aria-label="Toggle section">
	<i class="bi bi-chevron-right ms-2"></i>
	</a>
	</div>
	<ul id="quarto-sidebar-section-3" class="collapse list-unstyled sidebar-section depth2 show">
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/rag_zephyr_langchain.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">RAG Zephyr & LangChain</span></a>
	</div>
	</li>
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/advanced_rag.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">Advanced RAG</span></a>
	</div>
	</li>
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/rag_evaluation.html" class="sidebar-item-text sidebar-link active">
	<span class="menu-text">RAG Evaluation</span></a>
	</div>
	</li>
	</ul>
	</li>
	<li class="sidebar-item sidebar-item-section">
	<div class="sidebar-item-container">
	<a class="sidebar-item-text sidebar-link text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true">
	<span class="menu-text">Additional Techniques</span></a>
	<a class="sidebar-item-toggle text-start" data-bs-toggle="collapse" data-bs-target="#quarto-sidebar-section-4" role="navigation" aria-expanded="true" aria-label="Toggle section">
	<i class="bi bi-chevron-right ms-2"></i>
	</a>
	</div>
	<ul id="quarto-sidebar-section-4" class="collapse list-unstyled sidebar-section depth2 show">
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/automatic_embedding.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">Automatic Embedding</span></a>
	</div>
	</li>
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/faiss.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">FAISS for Efficient Search</span></a>
	</div>
	</li>
	<li class="sidebar-item">
	<div class="sidebar-item-container">
	<a href="../notebooks/single_gpu.html" class="sidebar-item-text sidebar-link">
	<span class="menu-text">Single GPU Optimization</span></a>
	</div>
	</li>
	</ul>
	</li>
	</ul>
	</li>
	</ul>
	</div>
	</nav>
	<div id="quarto-sidebar-glass" class="quarto-sidebar-collapse-item" data-bs-toggle="collapse" data-bs-target=".quarto-sidebar-collapse-item"></div>
	<!-- margin-sidebar -->
	<div id="quarto-margin-sidebar" class="sidebar margin-sidebar">
	<nav id="TOC" role="doc-toc" class="toc-active">
	<h2 id="toc-title">On this page</h2>

	<ul>
	<li><a href="#load-your-knowledge-base" id="toc-load-your-knowledge-base" class="nav-link active" data-scroll-target="#load-your-knowledge-base">Load your knowledge base</a></li>
	<li><a href="#build-a-synthetic-dataset-for-evaluation" id="toc-build-a-synthetic-dataset-for-evaluation" class="nav-link" data-scroll-target="#build-a-synthetic-dataset-for-evaluation">1. Build a synthetic dataset for evaluation</a>
	<ul class="collapse">
	<li><a href="#prepare-source-documents" id="toc-prepare-source-documents" class="nav-link" data-scroll-target="#prepare-source-documents">1.1. Prepare source documents</a></li>
	<li><a href="#setup-agents-for-question-generation" id="toc-setup-agents-for-question-generation" class="nav-link" data-scroll-target="#setup-agents-for-question-generation">1.2. Setup agents for question generation</a></li>
	<li><a href="#setup-critique-agents" id="toc-setup-critique-agents" class="nav-link" data-scroll-target="#setup-critique-agents">1.3. Setup critique agents</a></li>
	</ul></li>
	<li><a href="#build-our-rag-system" id="toc-build-our-rag-system" class="nav-link" data-scroll-target="#build-our-rag-system">2. Build our RAG System</a>
	<ul class="collapse">
	<li><a href="#preprocessing-documents-to-build-our-vector-database" id="toc-preprocessing-documents-to-build-our-vector-database" class="nav-link" data-scroll-target="#preprocessing-documents-to-build-our-vector-database">2.1. Preprocessing documents to build our vector database</a></li>
	<li><a href="#retriever---embeddings" id="toc-retriever---embeddings" class="nav-link" data-scroll-target="#retriever---embeddings">2.2. Retriever - embeddings 🗂️</a></li>
	<li><a href="#reader---llm" id="toc-reader---llm" class="nav-link" data-scroll-target="#reader---llm">2.3. Reader - LLM 💬</a></li>
	</ul></li>
	<li><a href="#benchmarking-the-rag-system" id="toc-benchmarking-the-rag-system" class="nav-link" data-scroll-target="#benchmarking-the-rag-system">3. Benchmarking the RAG system</a>
	<ul class="collapse">
	<li><a href="#inspect-results" id="toc-inspect-results" class="nav-link" data-scroll-target="#inspect-results">Inspect results</a></li>
	<li><a href="#example-results" id="toc-example-results" class="nav-link" data-scroll-target="#example-results">Example results</a></li>
	</ul></li>
	</ul>
	</nav>
	</div>
	<!-- main -->
	<main class="content" id="quarto-document-content">

	<header id="title-block-header" class="quarto-title-block default"><nav class="quarto-page-breadcrumbs quarto-title-breadcrumbs d-none d-lg-block" aria-label="breadcrumb"><ol class="breadcrumb"><li class="breadcrumb-item">Open-Source AI Cookbook</li><li class="breadcrumb-item"><a href="../notebooks/rag_zephyr_langchain.html">RAG Techniques</a></li><li class="breadcrumb-item"><a href="../notebooks/rag_evaluation.html">RAG Evaluation</a></li></ol></nav>
	<div class="quarto-title">
	<h1 class="title">RAG Evaluation</h1>
	</div>



	<div class="quarto-title-meta">




	</div>



	</header>


	<div id="22784221" class="cell" data-execution_count="1">
	<div class="sourceCode cell-code" id="cb1"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="op">!</span>pip install <span class="op">-</span>q torch transformers transformers langchain sentence<span class="op">-</span>transformers faiss<span class="op">-</span>gpu openpyxl openai</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="01b9dea1" class="cell" data-execution_count="2">
	<div class="sourceCode cell-code" id="cb2"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb2-1"><a href="#cb2-1" aria-hidden="true" tabindex="-1"></a><span class="op">%</span>reload_ext autoreload</span>
	<span id="cb2-2"><a href="#cb2-2" aria-hidden="true" tabindex="-1"></a><span class="op">%</span>autoreload <span class="dv">2</span></span>
	<span id="cb2-3"><a href="#cb2-3" aria-hidden="true" tabindex="-1"></a><span class="op">%</span>reload_ext dotenv</span>
	<span id="cb2-4"><a href="#cb2-4" aria-hidden="true" tabindex="-1"></a><span class="op">%</span>dotenv</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="b2fdd362" class="cell" data-execution_count="3">
	<div class="sourceCode cell-code" id="cb3"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> tqdm.notebook <span class="im">import</span> tqdm</span>
	<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> pandas <span class="im">as</span> pd</span>
	<span id="cb3-3"><a href="#cb3-3" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> typing <span class="im">import</span> Optional, List, Tuple</span>
	<span id="cb3-4"><a href="#cb3-4" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_core.language_models <span class="im">import</span> BaseChatModel</span>
	<span id="cb3-5"><a href="#cb3-5" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> json</span>
	<span id="cb3-6"><a href="#cb3-6" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> datasets</span>
	<span id="cb3-7"><a href="#cb3-7" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb3-8"><a href="#cb3-8" aria-hidden="true" tabindex="-1"></a>pd.set_option(<span class="st">"display.max_colwidth"</span>, <span class="va">None</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<section id="load-your-knowledge-base" class="level3">
	<h3 class="anchored" data-anchor-id="load-your-knowledge-base">Load your knowledge base</h3>
	<div id="359836ac" class="cell" data-execution_count="4">
	<div class="sourceCode cell-code" id="cb4"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>ds <span class="op">=</span> datasets.load_dataset(<span class="st">"m-ric/huggingface_doc"</span>, split<span class="op">=</span><span class="st">"train"</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="build-a-synthetic-dataset-for-evaluation" class="level1">
	<h1>1. Build a synthetic dataset for evaluation</h1>
	<p>We first build a synthetic dataset of questions and associated contexts. The method is to get elements from our knowledge base, and ask an LLM to generate questions based on these documents.</p>
	<p>Then we setup other LLM agents to act as quality filters for the generated QA couples: each of them will act as the filter for a specific flaw.</p>
	<section id="prepare-source-documents" class="level3">
	<h3 class="anchored" data-anchor-id="prepare-source-documents">1.1. Prepare source documents</h3>
	<div id="36f649ad" class="cell" data-execution_count="5">
	<div class="sourceCode cell-code" id="cb5"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb5-1"><a href="#cb5-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.text_splitter <span class="im">import</span> RecursiveCharacterTextSplitter</span>
	<span id="cb5-2"><a href="#cb5-2" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.docstore.document <span class="im">import</span> Document <span class="im">as</span> LangchainDocument</span>
	<span id="cb5-3"><a href="#cb5-3" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb5-4"><a href="#cb5-4" aria-hidden="true" tabindex="-1"></a>langchain_docs <span class="op">=</span> [</span>
	<span id="cb5-5"><a href="#cb5-5" aria-hidden="true" tabindex="-1"></a> LangchainDocument(page_content<span class="op">=</span>doc[<span class="st">"text"</span>], metadata<span class="op">=</span>{<span class="st">"source"</span>: doc[<span class="st">"source"</span>]})</span>
	<span id="cb5-6"><a href="#cb5-6" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> doc <span class="kw">in</span> tqdm(ds)</span>
	<span id="cb5-7"><a href="#cb5-7" aria-hidden="true" tabindex="-1"></a>]</span>
	<span id="cb5-8"><a href="#cb5-8" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb5-9"><a href="#cb5-9" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb5-10"><a href="#cb5-10" aria-hidden="true" tabindex="-1"></a>text_splitter <span class="op">=</span> RecursiveCharacterTextSplitter(</span>
	<span id="cb5-11"><a href="#cb5-11" aria-hidden="true" tabindex="-1"></a> chunk_size<span class="op">=</span><span class="dv">2000</span>,</span>
	<span id="cb5-12"><a href="#cb5-12" aria-hidden="true" tabindex="-1"></a> chunk_overlap<span class="op">=</span><span class="dv">200</span>,</span>
	<span id="cb5-13"><a href="#cb5-13" aria-hidden="true" tabindex="-1"></a> add_start_index<span class="op">=</span><span class="va">True</span>,</span>
	<span id="cb5-14"><a href="#cb5-14" aria-hidden="true" tabindex="-1"></a> separators<span class="op">=</span>[<span class="st">"</span><span class="ch">\n\n</span><span class="st">"</span>, <span class="st">"</span><span class="ch">\n</span><span class="st">"</span>, <span class="st">"."</span>, <span class="st">" "</span>, <span class="st">""</span>],</span>
	<span id="cb5-15"><a href="#cb5-15" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb5-16"><a href="#cb5-16" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb5-17"><a href="#cb5-17" aria-hidden="true" tabindex="-1"></a>docs_processed <span class="op">=</span> []</span>
	<span id="cb5-18"><a href="#cb5-18" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> doc <span class="kw">in</span> langchain_docs:</span>
	<span id="cb5-19"><a href="#cb5-19" aria-hidden="true" tabindex="-1"></a> docs_processed <span class="op">+=</span> text_splitter.split_documents([doc])</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="setup-agents-for-question-generation" class="level3">
	<h3 class="anchored" data-anchor-id="setup-agents-for-question-generation">1.2. Setup agents for question generation</h3>
	<p>We use <a href="https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1">Mixtral</a> for QA couple generation because it it has excellent performance in leaderboards such as <a href="https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard">Chatbot Arena</a>.</p>
	<div id="03878328" class="cell" data-execution_count="6">
	<div class="sourceCode cell-code" id="cb6"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_community.llms <span class="im">import</span> HuggingFaceHub</span>
	<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>repo_id <span class="op">=</span> <span class="st">"mistralai/Mixtral-8x7B-Instruct-v0.1"</span></span>
	<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>llm <span class="op">=</span> HuggingFaceHub(</span>
	<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a> repo_id<span class="op">=</span>repo_id,</span>
	<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a> task<span class="op">=</span><span class="st">"text-generation"</span>,</span>
	<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a> model_kwargs<span class="op">=</span>{</span>
	<span id="cb6-9"><a href="#cb6-9" aria-hidden="true" tabindex="-1"></a> <span class="st">"max_new_tokens"</span>: <span class="dv">512</span>,</span>
	<span id="cb6-10"><a href="#cb6-10" aria-hidden="true" tabindex="-1"></a> <span class="st">"top_k"</span>: <span class="dv">30</span>,</span>
	<span id="cb6-11"><a href="#cb6-11" aria-hidden="true" tabindex="-1"></a> <span class="st">"temperature"</span>: <span class="fl">0.1</span>,</span>
	<span id="cb6-12"><a href="#cb6-12" aria-hidden="true" tabindex="-1"></a> <span class="st">"repetition_penalty"</span>: <span class="fl">1.03</span>,</span>
	<span id="cb6-13"><a href="#cb6-13" aria-hidden="true" tabindex="-1"></a> },</span>
	<span id="cb6-14"><a href="#cb6-14" aria-hidden="true" tabindex="-1"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="67115f45" class="cell" data-execution_count="7">
	<div class="sourceCode cell-code" id="cb7"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_community.chat_models <span class="im">import</span> ChatHuggingFace</span>
	<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>chat_model <span class="op">=</span> ChatHuggingFace(llm<span class="op">=</span>llm)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="8a75ff72" class="cell" data-execution_count="8">
	<div class="sourceCode cell-code" id="cb8"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb8-1"><a href="#cb8-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.prompts <span class="im">import</span> ChatPromptTemplate</span>
	<span id="cb8-2"><a href="#cb8-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-3"><a href="#cb8-3" aria-hidden="true" tabindex="-1"></a>QA_generation_prompt <span class="op">=</span> <span class="st">"""</span></span>
	<span id="cb8-4"><a href="#cb8-4" aria-hidden="true" tabindex="-1"></a><span class="st">Your task is to write a factoid question and an answer given a context.</span></span>
	<span id="cb8-5"><a href="#cb8-5" aria-hidden="true" tabindex="-1"></a><span class="st">Your factoid question should be answerable with a specific, concise piece of factual information from the context.</span></span>
	<span id="cb8-6"><a href="#cb8-6" aria-hidden="true" tabindex="-1"></a><span class="st">Your factoid question should be formulated in the same style as questions users could ask in a search engine.</span></span>
	<span id="cb8-7"><a href="#cb8-7" aria-hidden="true" tabindex="-1"></a><span class="st">This means that your factoid question MUST NOT mention something like "according to the passage" or "context".</span></span>
	<span id="cb8-8"><a href="#cb8-8" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-9"><a href="#cb8-9" aria-hidden="true" tabindex="-1"></a><span class="st">Provide your answer as follows:</span></span>
	<span id="cb8-10"><a href="#cb8-10" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-11"><a href="#cb8-11" aria-hidden="true" tabindex="-1"></a><span class="st">Output:::</span></span>
	<span id="cb8-12"><a href="#cb8-12" aria-hidden="true" tabindex="-1"></a><span class="st">Factoid question: (your factoid question)</span></span>
	<span id="cb8-13"><a href="#cb8-13" aria-hidden="true" tabindex="-1"></a><span class="st">Answer: (your answer to the factoid question)</span></span>
	<span id="cb8-14"><a href="#cb8-14" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-15"><a href="#cb8-15" aria-hidden="true" tabindex="-1"></a><span class="st">Now here is the context.</span></span>
	<span id="cb8-16"><a href="#cb8-16" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-17"><a href="#cb8-17" aria-hidden="true" tabindex="-1"></a><span class="st">Context: </span><span class="sc">{context}</span><span class="ch">\n</span></span>
	<span id="cb8-18"><a href="#cb8-18" aria-hidden="true" tabindex="-1"></a><span class="st">Output:::"""</span></span>
	<span id="cb8-19"><a href="#cb8-19" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb8-20"><a href="#cb8-20" aria-hidden="true" tabindex="-1"></a>QA_generation_prompt <span class="op">=</span> ChatPromptTemplate.from_template(QA_generation_prompt)</span>
	<span id="cb8-21"><a href="#cb8-21" aria-hidden="true" tabindex="-1"></a>QA_generation_agent <span class="op">=</span> QA_generation_prompt <span class="op">\|</span> chat_model</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<p>Now let’s generate our QA couples. For this example, we generate only 10 QA couples and will load the rest from the Hub.</p>
	<p>But for your specific knowledge base, given that you want to get at least ~100 test samples, and accounting for the fact that we will filter out around half of these with our critique agents later on, you should generate much more, in the >200 samples.</p>
	<div id="66448027" class="cell" data-execution_count="9">
	<div class="sourceCode cell-code" id="cb9"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb9-1"><a href="#cb9-1" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> random</span>
	<span id="cb9-2"><a href="#cb9-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb9-3"><a href="#cb9-3" aria-hidden="true" tabindex="-1"></a>N_GENERATIONS <span class="op">=</span> (</span>
	<span id="cb9-4"><a href="#cb9-4" aria-hidden="true" tabindex="-1"></a> <span class="dv">10</span> <span class="co"># We intentionally generate only 10 QA couples here for cost and time considerations</span></span>
	<span id="cb9-5"><a href="#cb9-5" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb9-6"><a href="#cb9-6" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb9-7"><a href="#cb9-7" aria-hidden="true" tabindex="-1"></a><span class="bu">print</span>(<span class="ss">f"Generating </span><span class="sc">{</span>N_GENERATIONS<span class="sc">}</span><span class="ss"> QA couples..."</span>)</span>
	<span id="cb9-8"><a href="#cb9-8" aria-hidden="true" tabindex="-1"></a>outputs <span class="op">=</span> []</span>
	<span id="cb9-9"><a href="#cb9-9" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> context <span class="kw">in</span> tqdm(random.sample(langchain_docs, N_GENERATIONS)):</span>
	<span id="cb9-10"><a href="#cb9-10" aria-hidden="true" tabindex="-1"></a> <span class="co"># Generate QA couple</span></span>
	<span id="cb9-11"><a href="#cb9-11" aria-hidden="true" tabindex="-1"></a> output_QA_couple <span class="op">=</span> QA_generation_agent.invoke({<span class="st">"context"</span>: context.page_content}).content</span>
	<span id="cb9-12"><a href="#cb9-12" aria-hidden="true" tabindex="-1"></a> <span class="cf">try</span>:</span>
	<span id="cb9-13"><a href="#cb9-13" aria-hidden="true" tabindex="-1"></a> question <span class="op">=</span> output_QA_couple.split(<span class="st">"Factoid question: "</span>)[<span class="dv">1</span>].split(<span class="st">"Answer: "</span>)[<span class="dv">0</span>]</span>
	<span id="cb9-14"><a href="#cb9-14" aria-hidden="true" tabindex="-1"></a> answer <span class="op">=</span> output_QA_couple.split(<span class="st">"Answer: "</span>)[<span class="dv">1</span>]</span>
	<span id="cb9-15"><a href="#cb9-15" aria-hidden="true" tabindex="-1"></a> outputs.append(</span>
	<span id="cb9-16"><a href="#cb9-16" aria-hidden="true" tabindex="-1"></a> {</span>
	<span id="cb9-17"><a href="#cb9-17" aria-hidden="true" tabindex="-1"></a> <span class="st">"context"</span>: context.page_content,</span>
	<span id="cb9-18"><a href="#cb9-18" aria-hidden="true" tabindex="-1"></a> <span class="st">"question"</span>: question,</span>
	<span id="cb9-19"><a href="#cb9-19" aria-hidden="true" tabindex="-1"></a> <span class="st">"answer"</span>: answer,</span>
	<span id="cb9-20"><a href="#cb9-20" aria-hidden="true" tabindex="-1"></a> <span class="st">"source_doc"</span>: context.metadata[<span class="st">"source"</span>],</span>
	<span id="cb9-21"><a href="#cb9-21" aria-hidden="true" tabindex="-1"></a> }</span>
	<span id="cb9-22"><a href="#cb9-22" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb9-23"><a href="#cb9-23" aria-hidden="true" tabindex="-1"></a> <span class="cf">except</span>:</span>
	<span id="cb9-24"><a href="#cb9-24" aria-hidden="true" tabindex="-1"></a> <span class="cf">continue</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="38586e3e" class="cell" data-execution_count="10">
	<div class="sourceCode cell-code" id="cb10"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>display(pd.DataFrame(outputs).head(<span class="dv">1</span>))</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="setup-critique-agents" class="level3">
	<h3 class="anchored" data-anchor-id="setup-critique-agents">1.3. Setup critique agents</h3>
	<p>The questions generated by the previous agent can have many flaws: we should do a quality check before validating these questions.</p>
	<p>We thus build critique agents that will rate each question on several criteria, given in <a href="https://huggingface.co/papers/2312.10003">this paper</a>: - <strong>Groundedness:</strong> can the question be answered from the given context? - <strong>Relevance:</strong> is the question relevant to users? For instance, <code>"What is the date when transformers 4.29.1 was released?"</code> is not relevant for ML practicioners.</p>
	<p>One last failure case we’ve noticed is when a function is tailored for the particular setting where the question was generated, but undecipherable by itself, like <code>"What is the name of the function used in this guide?"</code>. We also build a critique agent for this criteria: - <strong>Stand-alone</strong>: is the question understandable free of any context, for someone with domain knowledge/Internet access? The opposite of this would be <code>What is the function used in this article?</code> for a question generated from a specific blog article.</p>
	<p>We systematically score functions with all these agents, and whenever the score is too low for any one of the agents, we eliminate the question from our eval dataset.</p>
	<p>💡 <strong><em>When asking the agents to output a score, we first ask them to produce its rationale. This will help us verify scores, but most importantly, asking it to first output rationale gives the model more tokens to think and elaborate an answer before summarizing it into a single score token.</em></strong></p>
	<p>We now build and run these critique agents.</p>
	<div id="36f64eeb" class="cell" data-execution_count="11">
	<div class="sourceCode cell-code" id="cb11"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>question_groundedness_critique_prompt <span class="op">=</span> <span class="st">"""</span></span>
	<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a><span class="st">You will be given a context and a question.</span></span>
	<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a><span class="st">Your task is to provide a 'total rating' scoring how well one can answer the given question unambiguously with the given context.</span></span>
	<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a><span class="st">Give your answer on a scale of 1 to 5, where 1 means that the question is not answerable at all given the context, and 5 means that the question is clearly and unambiguously answerable with the context.</span></span>
	<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a><span class="st">Provide your answer as follows:</span></span>
	<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a><span class="st">Answer:::</span></span>
	<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a><span class="st">Evaluation: (your rationale for the rating)</span></span>
	<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a><span class="st">Total rating: (your rating)</span></span>
	<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a><span class="st">Now here are the question and context.</span></span>
	<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-14"><a href="#cb11-14" aria-hidden="true" tabindex="-1"></a><span class="st">Question: </span><span class="sc">{question}</span><span class="ch">\n</span></span>
	<span id="cb11-15"><a href="#cb11-15" aria-hidden="true" tabindex="-1"></a><span class="st">Context: </span><span class="sc">{context}</span><span class="ch">\n</span></span>
	<span id="cb11-16"><a href="#cb11-16" aria-hidden="true" tabindex="-1"></a><span class="st">Answer::: """</span></span>
	<span id="cb11-17"><a href="#cb11-17" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-18"><a href="#cb11-18" aria-hidden="true" tabindex="-1"></a>question_relevance_critique_prompt <span class="op">=</span> <span class="st">"""</span></span>
	<span id="cb11-19"><a href="#cb11-19" aria-hidden="true" tabindex="-1"></a><span class="st">You will be given a question.</span></span>
	<span id="cb11-20"><a href="#cb11-20" aria-hidden="true" tabindex="-1"></a><span class="st">Your task is to provide a 'total rating' representing how useful this question can be to machine learning developers building NLP applications with the Hugging Face ecosystem.</span></span>
	<span id="cb11-21"><a href="#cb11-21" aria-hidden="true" tabindex="-1"></a><span class="st">Give your answer on a scale of 1 to 5, where 1 means that the question is not useful at all, and 5 means that the question is extremely useful.</span></span>
	<span id="cb11-22"><a href="#cb11-22" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-23"><a href="#cb11-23" aria-hidden="true" tabindex="-1"></a><span class="st">Provide your answer as follows:</span></span>
	<span id="cb11-24"><a href="#cb11-24" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-25"><a href="#cb11-25" aria-hidden="true" tabindex="-1"></a><span class="st">Answer:::</span></span>
	<span id="cb11-26"><a href="#cb11-26" aria-hidden="true" tabindex="-1"></a><span class="st">Evaluation: (your rationale for the rating)</span></span>
	<span id="cb11-27"><a href="#cb11-27" aria-hidden="true" tabindex="-1"></a><span class="st">Total rating: (your rating)</span></span>
	<span id="cb11-28"><a href="#cb11-28" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-29"><a href="#cb11-29" aria-hidden="true" tabindex="-1"></a><span class="st">Now here is the question.</span></span>
	<span id="cb11-30"><a href="#cb11-30" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-31"><a href="#cb11-31" aria-hidden="true" tabindex="-1"></a><span class="st">Question: </span><span class="sc">{question}</span><span class="ch">\n</span></span>
	<span id="cb11-32"><a href="#cb11-32" aria-hidden="true" tabindex="-1"></a><span class="st">Answer::: """</span></span>
	<span id="cb11-33"><a href="#cb11-33" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-34"><a href="#cb11-34" aria-hidden="true" tabindex="-1"></a>question_standalone_critique_prompt <span class="op">=</span> <span class="st">"""</span></span>
	<span id="cb11-35"><a href="#cb11-35" aria-hidden="true" tabindex="-1"></a><span class="st">You will be given a question.</span></span>
	<span id="cb11-36"><a href="#cb11-36" aria-hidden="true" tabindex="-1"></a><span class="st">Your task is to provide a 'total rating' representing how context-independant this question is.</span></span>
	<span id="cb11-37"><a href="#cb11-37" aria-hidden="true" tabindex="-1"></a><span class="st">Give your answer on a scale of 1 to 5, where 1 means that the question only makes sense in a specific context, and 5 means that the question makes sense by itself.</span></span>
	<span id="cb11-38"><a href="#cb11-38" aria-hidden="true" tabindex="-1"></a><span class="st">For instance, if the question refers to a particular setting, like 'in the context' or 'in the document', the rating must be 1.</span></span>
	<span id="cb11-39"><a href="#cb11-39" aria-hidden="true" tabindex="-1"></a><span class="st">The questions can contain obscure technical nouns or acronyms like Gradio, Hub, Hugging Face or Space and still be a 5: it must simply be clear to an operator with access to documentation what the question is about.</span></span>
	<span id="cb11-40"><a href="#cb11-40" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-41"><a href="#cb11-41" aria-hidden="true" tabindex="-1"></a><span class="st">Provide your answer as follows:</span></span>
	<span id="cb11-42"><a href="#cb11-42" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-43"><a href="#cb11-43" aria-hidden="true" tabindex="-1"></a><span class="st">Answer:::</span></span>
	<span id="cb11-44"><a href="#cb11-44" aria-hidden="true" tabindex="-1"></a><span class="st">Evaluation: (your rationale for the rating)</span></span>
	<span id="cb11-45"><a href="#cb11-45" aria-hidden="true" tabindex="-1"></a><span class="st">Total rating: (your rating)</span></span>
	<span id="cb11-46"><a href="#cb11-46" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-47"><a href="#cb11-47" aria-hidden="true" tabindex="-1"></a><span class="st">Now here is the question.</span></span>
	<span id="cb11-48"><a href="#cb11-48" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-49"><a href="#cb11-49" aria-hidden="true" tabindex="-1"></a><span class="st">Question: </span><span class="sc">{question}</span><span class="ch">\n</span></span>
	<span id="cb11-50"><a href="#cb11-50" aria-hidden="true" tabindex="-1"></a><span class="st">Answer::: """</span></span>
	<span id="cb11-51"><a href="#cb11-51" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-52"><a href="#cb11-52" aria-hidden="true" tabindex="-1"></a>question_groundedness_critique_prompt <span class="op">=</span> ChatPromptTemplate.from_template(</span>
	<span id="cb11-53"><a href="#cb11-53" aria-hidden="true" tabindex="-1"></a> question_groundedness_critique_prompt</span>
	<span id="cb11-54"><a href="#cb11-54" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb11-55"><a href="#cb11-55" aria-hidden="true" tabindex="-1"></a>question_groundedness_critique_agent <span class="op">=</span> question_groundedness_critique_prompt <span class="op">\|</span> chat_model</span>
	<span id="cb11-56"><a href="#cb11-56" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-57"><a href="#cb11-57" aria-hidden="true" tabindex="-1"></a>question_relevance_critique_prompt <span class="op">=</span> ChatPromptTemplate.from_template(</span>
	<span id="cb11-58"><a href="#cb11-58" aria-hidden="true" tabindex="-1"></a> question_relevance_critique_prompt</span>
	<span id="cb11-59"><a href="#cb11-59" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb11-60"><a href="#cb11-60" aria-hidden="true" tabindex="-1"></a>question_relevance_critique_agent <span class="op">=</span> question_relevance_critique_prompt <span class="op">\|</span> chat_model</span>
	<span id="cb11-61"><a href="#cb11-61" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb11-62"><a href="#cb11-62" aria-hidden="true" tabindex="-1"></a>question_standalone_critique_prompt <span class="op">=</span> ChatPromptTemplate.from_template(</span>
	<span id="cb11-63"><a href="#cb11-63" aria-hidden="true" tabindex="-1"></a> question_standalone_critique_prompt</span>
	<span id="cb11-64"><a href="#cb11-64" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb11-65"><a href="#cb11-65" aria-hidden="true" tabindex="-1"></a>question_standalone_critique_agent <span class="op">=</span> question_standalone_critique_prompt <span class="op">\|</span> chat_model</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="36a9f0a0" class="cell" data-execution_count="12">
	<div class="sourceCode cell-code" id="cb12"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb12-1"><a href="#cb12-1" aria-hidden="true" tabindex="-1"></a><span class="bu">print</span>(<span class="st">"Generating critique for each QA couple..."</span>)</span>
	<span id="cb12-2"><a href="#cb12-2" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> output <span class="kw">in</span> tqdm(outputs):</span>
	<span id="cb12-3"><a href="#cb12-3" aria-hidden="true" tabindex="-1"></a> <span class="co"># Critique the generated QA couple</span></span>
	<span id="cb12-4"><a href="#cb12-4" aria-hidden="true" tabindex="-1"></a> question_groundedness_evaluation <span class="op">=</span> question_groundedness_critique_agent.invoke(</span>
	<span id="cb12-5"><a href="#cb12-5" aria-hidden="true" tabindex="-1"></a> {<span class="st">"context"</span>: output[<span class="st">"context"</span>], <span class="st">"question"</span>: output[<span class="st">"question"</span>]}</span>
	<span id="cb12-6"><a href="#cb12-6" aria-hidden="true" tabindex="-1"></a> ).content</span>
	<span id="cb12-7"><a href="#cb12-7" aria-hidden="true" tabindex="-1"></a> question_relevance_evaluation <span class="op">=</span> question_relevance_critique_agent.invoke(</span>
	<span id="cb12-8"><a href="#cb12-8" aria-hidden="true" tabindex="-1"></a> {<span class="st">"question"</span>: output[<span class="st">"question"</span>]}</span>
	<span id="cb12-9"><a href="#cb12-9" aria-hidden="true" tabindex="-1"></a> ).content</span>
	<span id="cb12-10"><a href="#cb12-10" aria-hidden="true" tabindex="-1"></a> question_standalone_evaluation <span class="op">=</span> question_standalone_critique_agent.invoke(</span>
	<span id="cb12-11"><a href="#cb12-11" aria-hidden="true" tabindex="-1"></a> {<span class="st">"question"</span>: output[<span class="st">"question"</span>]}</span>
	<span id="cb12-12"><a href="#cb12-12" aria-hidden="true" tabindex="-1"></a> ).content</span>
	<span id="cb12-13"><a href="#cb12-13" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb12-14"><a href="#cb12-14" aria-hidden="true" tabindex="-1"></a> <span class="cf">try</span>:</span>
	<span id="cb12-15"><a href="#cb12-15" aria-hidden="true" tabindex="-1"></a> groundedness_score <span class="op">=</span> <span class="bu">int</span>(question_groundedness_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">1</span>][<span class="dv">0</span>])</span>
	<span id="cb12-16"><a href="#cb12-16" aria-hidden="true" tabindex="-1"></a> groundedness_eval <span class="op">=</span> question_groundedness_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">0</span>].split(</span>
	<span id="cb12-17"><a href="#cb12-17" aria-hidden="true" tabindex="-1"></a> <span class="st">"Evaluation: "</span></span>
	<span id="cb12-18"><a href="#cb12-18" aria-hidden="true" tabindex="-1"></a> )[<span class="dv">1</span>]</span>
	<span id="cb12-19"><a href="#cb12-19" aria-hidden="true" tabindex="-1"></a> relevance_score <span class="op">=</span> <span class="bu">int</span>(question_relevance_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">1</span>][<span class="dv">0</span>])</span>
	<span id="cb12-20"><a href="#cb12-20" aria-hidden="true" tabindex="-1"></a> relevance_eval <span class="op">=</span> question_relevance_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">0</span>].split(</span>
	<span id="cb12-21"><a href="#cb12-21" aria-hidden="true" tabindex="-1"></a> <span class="st">"Evaluation: "</span></span>
	<span id="cb12-22"><a href="#cb12-22" aria-hidden="true" tabindex="-1"></a> )[<span class="dv">1</span>]</span>
	<span id="cb12-23"><a href="#cb12-23" aria-hidden="true" tabindex="-1"></a> standalone_score <span class="op">=</span> <span class="bu">int</span>(question_standalone_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">1</span>][<span class="dv">0</span>])</span>
	<span id="cb12-24"><a href="#cb12-24" aria-hidden="true" tabindex="-1"></a> standalone_eval <span class="op">=</span> question_standalone_evaluation.split(<span class="st">"Total rating: "</span>)[<span class="dv">0</span>].split(</span>
	<span id="cb12-25"><a href="#cb12-25" aria-hidden="true" tabindex="-1"></a> <span class="st">"Evaluation: "</span></span>
	<span id="cb12-26"><a href="#cb12-26" aria-hidden="true" tabindex="-1"></a> )[<span class="dv">1</span>]</span>
	<span id="cb12-27"><a href="#cb12-27" aria-hidden="true" tabindex="-1"></a> output.update(</span>
	<span id="cb12-28"><a href="#cb12-28" aria-hidden="true" tabindex="-1"></a> {</span>
	<span id="cb12-29"><a href="#cb12-29" aria-hidden="true" tabindex="-1"></a> <span class="st">"groundedness_score"</span>: groundedness_score,</span>
	<span id="cb12-30"><a href="#cb12-30" aria-hidden="true" tabindex="-1"></a> <span class="st">"groundedness_eval"</span>: groundedness_eval,</span>
	<span id="cb12-31"><a href="#cb12-31" aria-hidden="true" tabindex="-1"></a> <span class="st">"relevance_score"</span>: relevance_score,</span>
	<span id="cb12-32"><a href="#cb12-32" aria-hidden="true" tabindex="-1"></a> <span class="st">"relevance_eval"</span>: relevance_eval,</span>
	<span id="cb12-33"><a href="#cb12-33" aria-hidden="true" tabindex="-1"></a> <span class="st">"standalone_score"</span>: standalone_score,</span>
	<span id="cb12-34"><a href="#cb12-34" aria-hidden="true" tabindex="-1"></a> <span class="st">"standalone_eval"</span>: standalone_eval,</span>
	<span id="cb12-35"><a href="#cb12-35" aria-hidden="true" tabindex="-1"></a> }</span>
	<span id="cb12-36"><a href="#cb12-36" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb12-37"><a href="#cb12-37" aria-hidden="true" tabindex="-1"></a> <span class="cf">except</span>:</span>
	<span id="cb12-38"><a href="#cb12-38" aria-hidden="true" tabindex="-1"></a> <span class="cf">continue</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<p>Now let us filter out bad questions based on our critique agent scores:</p>
	<div id="244dd1b5" class="cell" data-execution_count="13">
	<div class="sourceCode cell-code" id="cb13"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> pandas <span class="im">as</span> pd</span>
	<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb13-3"><a href="#cb13-3" aria-hidden="true" tabindex="-1"></a>pd.set_option(<span class="st">"display.max_colwidth"</span>, <span class="va">None</span>)</span>
	<span id="cb13-4"><a href="#cb13-4" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb13-5"><a href="#cb13-5" aria-hidden="true" tabindex="-1"></a>generated_questions <span class="op">=</span> pd.DataFrame.from_dict(outputs)</span>
	<span id="cb13-6"><a href="#cb13-6" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb13-7"><a href="#cb13-7" aria-hidden="true" tabindex="-1"></a><span class="bu">print</span>(<span class="st">"Evaluation dataset before filtering:"</span>)</span>
	<span id="cb13-8"><a href="#cb13-8" aria-hidden="true" tabindex="-1"></a>display(</span>
	<span id="cb13-9"><a href="#cb13-9" aria-hidden="true" tabindex="-1"></a> generated_questions[</span>
	<span id="cb13-10"><a href="#cb13-10" aria-hidden="true" tabindex="-1"></a> [<span class="st">"question"</span>, <span class="st">"answer"</span>, <span class="st">"groundedness_score"</span>, <span class="st">"relevance_score"</span>, <span class="st">"standalone_score"</span>]</span>
	<span id="cb13-11"><a href="#cb13-11" aria-hidden="true" tabindex="-1"></a> ]</span>
	<span id="cb13-12"><a href="#cb13-12" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb13-13"><a href="#cb13-13" aria-hidden="true" tabindex="-1"></a>generated_questions <span class="op">=</span> generated_questions.loc[</span>
	<span id="cb13-14"><a href="#cb13-14" aria-hidden="true" tabindex="-1"></a> (generated_questions[<span class="st">"groundedness_score"</span>] <span class="op">>=</span> <span class="dv">4</span>)</span>
	<span id="cb13-15"><a href="#cb13-15" aria-hidden="true" tabindex="-1"></a> <span class="op">&</span> (generated_questions[<span class="st">"relevance_score"</span>] <span class="op">>=</span> <span class="dv">4</span>)</span>
	<span id="cb13-16"><a href="#cb13-16" aria-hidden="true" tabindex="-1"></a> <span class="op">&</span> (generated_questions[<span class="st">"standalone_score"</span>] <span class="op">>=</span> <span class="dv">4</span>)</span>
	<span id="cb13-17"><a href="#cb13-17" aria-hidden="true" tabindex="-1"></a>]</span>
	<span id="cb13-18"><a href="#cb13-18" aria-hidden="true" tabindex="-1"></a><span class="bu">print</span>(<span class="st">"============================================"</span>)</span>
	<span id="cb13-19"><a href="#cb13-19" aria-hidden="true" tabindex="-1"></a><span class="bu">print</span>(<span class="st">"Final evaluation dataset:"</span>)</span>
	<span id="cb13-20"><a href="#cb13-20" aria-hidden="true" tabindex="-1"></a>display(</span>
	<span id="cb13-21"><a href="#cb13-21" aria-hidden="true" tabindex="-1"></a> generated_questions[</span>
	<span id="cb13-22"><a href="#cb13-22" aria-hidden="true" tabindex="-1"></a> [<span class="st">"question"</span>, <span class="st">"answer"</span>, <span class="st">"groundedness_score"</span>, <span class="st">"relevance_score"</span>, <span class="st">"standalone_score"</span>]</span>
	<span id="cb13-23"><a href="#cb13-23" aria-hidden="true" tabindex="-1"></a> ]</span>
	<span id="cb13-24"><a href="#cb13-24" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb13-25"><a href="#cb13-25" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb13-26"><a href="#cb13-26" aria-hidden="true" tabindex="-1"></a>eval_dataset <span class="op">=</span> datasets.Dataset.from_pandas(</span>
	<span id="cb13-27"><a href="#cb13-27" aria-hidden="true" tabindex="-1"></a> generated_questions, split<span class="op">=</span><span class="st">"train"</span>, preserve_index<span class="op">=</span><span class="va">False</span></span>
	<span id="cb13-28"><a href="#cb13-28" aria-hidden="true" tabindex="-1"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<p>Now our synthetic evaluation dataset is complete! We can evaluate different RAG systems on this evaluation dataset.</p>
	<p>We have generated only a few QA couples here to reduce time and cost. But let’s kick start the next part by loading a pre-generated dataset:</p>
	<div id="9e9d2a9f" class="cell" data-execution_count="14">
	<div class="sourceCode cell-code" id="cb14"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb14-1"><a href="#cb14-1" aria-hidden="true" tabindex="-1"></a>eval_dataset <span class="op">=</span> datasets.load_dataset(<span class="st">"m-ric/huggingface_doc_qa_eval"</span>, split<span class="op">=</span><span class="st">"train"</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	</section>
	<section id="build-our-rag-system" class="level1">
	<h1>2. Build our RAG System</h1>
	<section id="preprocessing-documents-to-build-our-vector-database" class="level3">
	<h3 class="anchored" data-anchor-id="preprocessing-documents-to-build-our-vector-database">2.1. Preprocessing documents to build our vector database</h3>
	<ul>
	<li>In this part, <strong>we split the documents from our knowledge base into smaller chunks</strong>: these will be the snippets that are picked by the Retriever, to then be ingested by the Reader LLM as supporting elements for its answer.</li>
	<li>The goal is to build semantically relevant snippets: not too small to be sufficient for supporting an answer, and not too large too avoid diluting individual ideas.</li>
	</ul>
	<p>Many options exist for text splitting: - split every <code>n</code> words / characters, but this has the risk of cutting in half paragraphs or even sentences - split after <code>n</code> words / character, but only on sentence boundaries - <strong>recursive split</strong> tries to preserve even more of the document structure, by processing it tree-like way, splitting first on the largest units (chapters) then recursively splitting on smaller units (paragraphs, sentences).</p>
	<p>To learn more about chunking, I recommend you read <a href="https://github.com/FullStackRetrieval-com/RetrievalTutorials/blob/main/5_Levels_Of_Text_Splitting.ipynb">this great notebook</a> by Greg Kamradt.</p>
	<p><a href="https://huggingface.co/spaces/m-ric/chunk_visualizer">This space</a> lets you visualize how different splitting options affect the chunks you get.</p>
	<blockquote class="blockquote">
	<p>In the following, we use Langchain’s <code>RecursiveCharacterTextSplitter</code>.</p>
	</blockquote>
	<p>💡 <em>To measure chunk length in our Text Splitter, our length function will not be the count of characters, but the count of tokens in the tokenized text: indeed, for subsequent embedder that processes token, measuring length in tokens is more relevant and empirically performs better.</em></p>
	<div id="94af026d" class="cell" data-execution_count="15">
	<div class="sourceCode cell-code" id="cb15"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb15-1"><a href="#cb15-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.docstore.document <span class="im">import</span> Document <span class="im">as</span> LangchainDocument</span>
	<span id="cb15-2"><a href="#cb15-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb15-3"><a href="#cb15-3" aria-hidden="true" tabindex="-1"></a>RAW_KNOWLEDGE_BASE <span class="op">=</span> [</span>
	<span id="cb15-4"><a href="#cb15-4" aria-hidden="true" tabindex="-1"></a> LangchainDocument(page_content<span class="op">=</span>doc[<span class="st">"text"</span>], metadata<span class="op">=</span>{<span class="st">"source"</span>: doc[<span class="st">"source"</span>]})</span>
	<span id="cb15-5"><a href="#cb15-5" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> doc <span class="kw">in</span> tqdm(ds)</span>
	<span id="cb15-6"><a href="#cb15-6" aria-hidden="true" tabindex="-1"></a>]</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="13cebd63" class="cell" data-execution_count="16">
	<div class="sourceCode cell-code" id="cb16"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.text_splitter <span class="im">import</span> RecursiveCharacterTextSplitter</span>
	<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> transformers <span class="im">import</span> AutoTokenizer</span>
	<span id="cb16-3"><a href="#cb16-3" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb16-4"><a href="#cb16-4" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb16-5"><a href="#cb16-5" aria-hidden="true" tabindex="-1"></a><span class="kw">def</span> split_documents(</span>
	<span id="cb16-6"><a href="#cb16-6" aria-hidden="true" tabindex="-1"></a> chunk_size: <span class="bu">int</span>,</span>
	<span id="cb16-7"><a href="#cb16-7" aria-hidden="true" tabindex="-1"></a> knowledge_base: List[LangchainDocument],</span>
	<span id="cb16-8"><a href="#cb16-8" aria-hidden="true" tabindex="-1"></a> tokenizer_name: <span class="bu">str</span>,</span>
	<span id="cb16-9"><a href="#cb16-9" aria-hidden="true" tabindex="-1"></a>) <span class="op">-></span> List[LangchainDocument]:</span>
	<span id="cb16-10"><a href="#cb16-10" aria-hidden="true" tabindex="-1"></a> <span class="co">"""</span></span>
	<span id="cb16-11"><a href="#cb16-11" aria-hidden="true" tabindex="-1"></a><span class="co"> Split documents into chunks of size `chunk_size` characters and return a list of documents.</span></span>
	<span id="cb16-12"><a href="#cb16-12" aria-hidden="true" tabindex="-1"></a><span class="co"> """</span></span>
	<span id="cb16-13"><a href="#cb16-13" aria-hidden="true" tabindex="-1"></a> text_splitter <span class="op">=</span> RecursiveCharacterTextSplitter.from_huggingface_tokenizer(</span>
	<span id="cb16-14"><a href="#cb16-14" aria-hidden="true" tabindex="-1"></a> AutoTokenizer.from_pretrained(tokenizer_name),</span>
	<span id="cb16-15"><a href="#cb16-15" aria-hidden="true" tabindex="-1"></a> chunk_size<span class="op">=</span>chunk_size,</span>
	<span id="cb16-16"><a href="#cb16-16" aria-hidden="true" tabindex="-1"></a> chunk_overlap<span class="op">=</span><span class="bu">int</span>(chunk_size <span class="op">/</span> <span class="dv">10</span>),</span>
	<span id="cb16-17"><a href="#cb16-17" aria-hidden="true" tabindex="-1"></a> add_start_index<span class="op">=</span><span class="va">True</span>,</span>
	<span id="cb16-18"><a href="#cb16-18" aria-hidden="true" tabindex="-1"></a> strip_whitespace<span class="op">=</span><span class="va">True</span>,</span>
	<span id="cb16-19"><a href="#cb16-19" aria-hidden="true" tabindex="-1"></a> separators<span class="op">=</span>[<span class="st">"</span><span class="ch">\n\n</span><span class="st">"</span>, <span class="st">"</span><span class="ch">\n</span><span class="st">"</span>, <span class="st">"."</span>, <span class="st">" "</span>, <span class="st">""</span>],</span>
	<span id="cb16-20"><a href="#cb16-20" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb16-21"><a href="#cb16-21" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb16-22"><a href="#cb16-22" aria-hidden="true" tabindex="-1"></a> docs_processed <span class="op">=</span> []</span>
	<span id="cb16-23"><a href="#cb16-23" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> doc <span class="kw">in</span> knowledge_base:</span>
	<span id="cb16-24"><a href="#cb16-24" aria-hidden="true" tabindex="-1"></a> docs_processed <span class="op">+=</span> text_splitter.split_documents([doc])</span>
	<span id="cb16-25"><a href="#cb16-25" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb16-26"><a href="#cb16-26" aria-hidden="true" tabindex="-1"></a> <span class="co"># Remove duplicates</span></span>
	<span id="cb16-27"><a href="#cb16-27" aria-hidden="true" tabindex="-1"></a> unique_texts <span class="op">=</span> {}</span>
	<span id="cb16-28"><a href="#cb16-28" aria-hidden="true" tabindex="-1"></a> docs_processed_unique <span class="op">=</span> []</span>
	<span id="cb16-29"><a href="#cb16-29" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> doc <span class="kw">in</span> docs_processed:</span>
	<span id="cb16-30"><a href="#cb16-30" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> doc.page_content <span class="kw">not</span> <span class="kw">in</span> unique_texts:</span>
	<span id="cb16-31"><a href="#cb16-31" aria-hidden="true" tabindex="-1"></a> unique_texts[doc.page_content] <span class="op">=</span> <span class="va">True</span></span>
	<span id="cb16-32"><a href="#cb16-32" aria-hidden="true" tabindex="-1"></a> docs_processed_unique.append(doc)</span>
	<span id="cb16-33"><a href="#cb16-33" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb16-34"><a href="#cb16-34" aria-hidden="true" tabindex="-1"></a> <span class="cf">return</span> docs_processed_unique</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="retriever---embeddings" class="level3">
	<h3 class="anchored" data-anchor-id="retriever---embeddings">2.2. Retriever - embeddings 🗂️</h3>
	<p>The <strong>retriever acts like an internal search engine</strong>: given the user query, it returns the most relevant documents from your knowledge base.</p>
	<blockquote class="blockquote">
	<p>For the knowledge base, we use Langchain vector databases since <strong>it offers a convenient <a href="https://github.com/facebookresearch/faiss">FAISS</a> index and allows us to keep document metadata throughout the processing</strong>.</p>
	</blockquote>
	<p>🛠️ <strong>Options included:</strong></p>
	<ul>
	<li>Tune the chunking method:
	<ul>
	<li>Size of the chunks</li>
	<li>Method: split on different separators, use <a href="https://python.langchain.com/docs/modules/data_connection/document_transformers/semantic-chunker">semantic chunking</a>…</li>
	</ul></li>
	<li>Change the embedding model</li>
	</ul>
	<div id="8f7371a4" class="cell" data-execution_count="17">
	<div class="sourceCode cell-code" id="cb17"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb17-1"><a href="#cb17-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.vectorstores <span class="im">import</span> FAISS</span>
	<span id="cb17-2"><a href="#cb17-2" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_community.embeddings <span class="im">import</span> HuggingFaceEmbeddings</span>
	<span id="cb17-3"><a href="#cb17-3" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_community.vectorstores.utils <span class="im">import</span> DistanceStrategy</span>
	<span id="cb17-4"><a href="#cb17-4" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> os</span>
	<span id="cb17-5"><a href="#cb17-5" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-6"><a href="#cb17-6" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-7"><a href="#cb17-7" aria-hidden="true" tabindex="-1"></a><span class="kw">def</span> load_embeddings(</span>
	<span id="cb17-8"><a href="#cb17-8" aria-hidden="true" tabindex="-1"></a> langchain_docs: List[LangchainDocument],</span>
	<span id="cb17-9"><a href="#cb17-9" aria-hidden="true" tabindex="-1"></a> chunk_size: <span class="bu">int</span>,</span>
	<span id="cb17-10"><a href="#cb17-10" aria-hidden="true" tabindex="-1"></a> embedding_model_name: Optional[<span class="bu">str</span>] <span class="op">=</span> <span class="st">"thenlper/gte-small"</span>,</span>
	<span id="cb17-11"><a href="#cb17-11" aria-hidden="true" tabindex="-1"></a>) <span class="op">-></span> FAISS:</span>
	<span id="cb17-12"><a href="#cb17-12" aria-hidden="true" tabindex="-1"></a> <span class="co">"""</span></span>
	<span id="cb17-13"><a href="#cb17-13" aria-hidden="true" tabindex="-1"></a><span class="co"> Creates a FAISS index from the given embedding model and documents. Loads the index directly if it already exists.</span></span>
	<span id="cb17-14"><a href="#cb17-14" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-15"><a href="#cb17-15" aria-hidden="true" tabindex="-1"></a><span class="co"> Args:</span></span>
	<span id="cb17-16"><a href="#cb17-16" aria-hidden="true" tabindex="-1"></a><span class="co"> langchain_docs: list of documents</span></span>
	<span id="cb17-17"><a href="#cb17-17" aria-hidden="true" tabindex="-1"></a><span class="co"> chunk_size: size of the chunks to split the documents into</span></span>
	<span id="cb17-18"><a href="#cb17-18" aria-hidden="true" tabindex="-1"></a><span class="co"> embedding_model_name: name of the embedding model to use</span></span>
	<span id="cb17-19"><a href="#cb17-19" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-20"><a href="#cb17-20" aria-hidden="true" tabindex="-1"></a><span class="co"> Returns:</span></span>
	<span id="cb17-21"><a href="#cb17-21" aria-hidden="true" tabindex="-1"></a><span class="co"> FAISS index</span></span>
	<span id="cb17-22"><a href="#cb17-22" aria-hidden="true" tabindex="-1"></a><span class="co"> """</span></span>
	<span id="cb17-23"><a href="#cb17-23" aria-hidden="true" tabindex="-1"></a> <span class="co"># load embedding_model</span></span>
	<span id="cb17-24"><a href="#cb17-24" aria-hidden="true" tabindex="-1"></a> embedding_model <span class="op">=</span> HuggingFaceEmbeddings(</span>
	<span id="cb17-25"><a href="#cb17-25" aria-hidden="true" tabindex="-1"></a> model_name<span class="op">=</span>embedding_model_name,</span>
	<span id="cb17-26"><a href="#cb17-26" aria-hidden="true" tabindex="-1"></a> multi_process<span class="op">=</span><span class="va">True</span>,</span>
	<span id="cb17-27"><a href="#cb17-27" aria-hidden="true" tabindex="-1"></a> model_kwargs<span class="op">=</span>{<span class="st">"device"</span>: <span class="st">"cuda"</span>},</span>
	<span id="cb17-28"><a href="#cb17-28" aria-hidden="true" tabindex="-1"></a> encode_kwargs<span class="op">=</span>{<span class="st">"normalize_embeddings"</span>: <span class="va">True</span>}, <span class="co"># set True to compute cosine similarity</span></span>
	<span id="cb17-29"><a href="#cb17-29" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb17-30"><a href="#cb17-30" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-31"><a href="#cb17-31" aria-hidden="true" tabindex="-1"></a> <span class="co"># Check if embeddings already exist on disk</span></span>
	<span id="cb17-32"><a href="#cb17-32" aria-hidden="true" tabindex="-1"></a> index_name <span class="op">=</span> <span class="ss">f"index_chunk:</span><span class="sc">{</span>chunk_size<span class="sc">}</span><span class="ss">_embeddings:</span><span class="sc">{</span>embedding_model_name<span class="sc">.</span>replace(<span class="st">'/'</span>, <span class="st">'~'</span>)<span class="sc">}</span><span class="ss">"</span></span>
	<span id="cb17-33"><a href="#cb17-33" aria-hidden="true" tabindex="-1"></a> index_folder_path <span class="op">=</span> <span class="ss">f"./data/indexes/</span><span class="sc">{</span>index_name<span class="sc">}</span><span class="ss">/"</span></span>
	<span id="cb17-34"><a href="#cb17-34" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> os.path.isdir(index_folder_path):</span>
	<span id="cb17-35"><a href="#cb17-35" aria-hidden="true" tabindex="-1"></a> <span class="cf">return</span> FAISS.load_local(</span>
	<span id="cb17-36"><a href="#cb17-36" aria-hidden="true" tabindex="-1"></a> index_folder_path,</span>
	<span id="cb17-37"><a href="#cb17-37" aria-hidden="true" tabindex="-1"></a> embedding_model,</span>
	<span id="cb17-38"><a href="#cb17-38" aria-hidden="true" tabindex="-1"></a> distance_strategy<span class="op">=</span>DistanceStrategy.COSINE,</span>
	<span id="cb17-39"><a href="#cb17-39" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb17-40"><a href="#cb17-40" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb17-41"><a href="#cb17-41" aria-hidden="true" tabindex="-1"></a> <span class="cf">else</span>:</span>
	<span id="cb17-42"><a href="#cb17-42" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="st">"Index not found, generating it..."</span>)</span>
	<span id="cb17-43"><a href="#cb17-43" aria-hidden="true" tabindex="-1"></a> docs_processed <span class="op">=</span> split_documents(</span>
	<span id="cb17-44"><a href="#cb17-44" aria-hidden="true" tabindex="-1"></a> chunk_size,</span>
	<span id="cb17-45"><a href="#cb17-45" aria-hidden="true" tabindex="-1"></a> langchain_docs,</span>
	<span id="cb17-46"><a href="#cb17-46" aria-hidden="true" tabindex="-1"></a> embedding_model_name,</span>
	<span id="cb17-47"><a href="#cb17-47" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb17-48"><a href="#cb17-48" aria-hidden="true" tabindex="-1"></a> knowledge_index <span class="op">=</span> FAISS.from_documents(</span>
	<span id="cb17-49"><a href="#cb17-49" aria-hidden="true" tabindex="-1"></a> docs_processed, embedding_model, distance_strategy<span class="op">=</span>DistanceStrategy.COSINE</span>
	<span id="cb17-50"><a href="#cb17-50" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb17-51"><a href="#cb17-51" aria-hidden="true" tabindex="-1"></a> knowledge_index.save_local(index_folder_path)</span>
	<span id="cb17-52"><a href="#cb17-52" aria-hidden="true" tabindex="-1"></a> <span class="cf">return</span> knowledge_index</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="reader---llm" class="level3">
	<h3 class="anchored" data-anchor-id="reader---llm">2.3. Reader - LLM 💬</h3>
	<p>In this part, the <strong>LLM Reader reads the retrieved documents to formulate its answer.</strong></p>
	<p>🛠️ Here we tried the following options to improve results: - Switch reranking on/off - Change the reader model</p>
	<div id="843d7987" class="cell" data-execution_count="18">
	<div class="sourceCode cell-code" id="cb18"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb18-1"><a href="#cb18-1" aria-hidden="true" tabindex="-1"></a>RAG_PROMPT_TEMPLATE <span class="op">=</span> <span class="st">"""</span></span>
	<span id="cb18-2"><a href="#cb18-2" aria-hidden="true" tabindex="-1"></a><span class="st"><\|system\|></span></span>
	<span id="cb18-3"><a href="#cb18-3" aria-hidden="true" tabindex="-1"></a><span class="st">Using the information contained in the context,</span></span>
	<span id="cb18-4"><a href="#cb18-4" aria-hidden="true" tabindex="-1"></a><span class="st">give a comprehensive answer to the question.</span></span>
	<span id="cb18-5"><a href="#cb18-5" aria-hidden="true" tabindex="-1"></a><span class="st">Respond only to the question asked, response should be concise and relevant to the question.</span></span>
	<span id="cb18-6"><a href="#cb18-6" aria-hidden="true" tabindex="-1"></a><span class="st">Provide the number of the source document when relevant.</span></span>
	<span id="cb18-7"><a href="#cb18-7" aria-hidden="true" tabindex="-1"></a><span class="st">If the answer cannot be deduced from the context, do not give an answer.</s></span></span>
	<span id="cb18-8"><a href="#cb18-8" aria-hidden="true" tabindex="-1"></a><span class="st"><\|user\|></span></span>
	<span id="cb18-9"><a href="#cb18-9" aria-hidden="true" tabindex="-1"></a><span class="st">Context:</span></span>
	<span id="cb18-10"><a href="#cb18-10" aria-hidden="true" tabindex="-1"></a><span class="sc">{context}</span></span>
	<span id="cb18-11"><a href="#cb18-11" aria-hidden="true" tabindex="-1"></a><span class="st">---</span></span>
	<span id="cb18-12"><a href="#cb18-12" aria-hidden="true" tabindex="-1"></a><span class="st">Now here is the question you need to answer.</span></span>
	<span id="cb18-13"><a href="#cb18-13" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb18-14"><a href="#cb18-14" aria-hidden="true" tabindex="-1"></a><span class="st">Question: </span><span class="sc">{question}</span></span>
	<span id="cb18-15"><a href="#cb18-15" aria-hidden="true" tabindex="-1"></a><span class="st"></s></span></span>
	<span id="cb18-16"><a href="#cb18-16" aria-hidden="true" tabindex="-1"></a><span class="st"><\|assistant\|></span></span>
	<span id="cb18-17"><a href="#cb18-17" aria-hidden="true" tabindex="-1"></a><span class="st">"""</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="6884550b" class="cell" data-execution_count="19">
	<div class="sourceCode cell-code" id="cb19"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_community.llms <span class="im">import</span> HuggingFaceHub</span>
	<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb19-3"><a href="#cb19-3" aria-hidden="true" tabindex="-1"></a>repo_id <span class="op">=</span> <span class="st">"HuggingFaceH4/zephyr-7b-beta"</span></span>
	<span id="cb19-4"><a href="#cb19-4" aria-hidden="true" tabindex="-1"></a>READER_MODEL_NAME <span class="op">=</span> <span class="st">"zephyr-7b-beta"</span></span>
	<span id="cb19-5"><a href="#cb19-5" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb19-6"><a href="#cb19-6" aria-hidden="true" tabindex="-1"></a>READER_LLM <span class="op">=</span> HuggingFaceHub(</span>
	<span id="cb19-7"><a href="#cb19-7" aria-hidden="true" tabindex="-1"></a> repo_id<span class="op">=</span>repo_id,</span>
	<span id="cb19-8"><a href="#cb19-8" aria-hidden="true" tabindex="-1"></a> task<span class="op">=</span><span class="st">"text-generation"</span>,</span>
	<span id="cb19-9"><a href="#cb19-9" aria-hidden="true" tabindex="-1"></a> model_kwargs<span class="op">=</span>{</span>
	<span id="cb19-10"><a href="#cb19-10" aria-hidden="true" tabindex="-1"></a> <span class="st">"max_new_tokens"</span>: <span class="dv">512</span>,</span>
	<span id="cb19-11"><a href="#cb19-11" aria-hidden="true" tabindex="-1"></a> <span class="st">"top_k"</span>: <span class="dv">30</span>,</span>
	<span id="cb19-12"><a href="#cb19-12" aria-hidden="true" tabindex="-1"></a> <span class="st">"temperature"</span>: <span class="fl">0.1</span>,</span>
	<span id="cb19-13"><a href="#cb19-13" aria-hidden="true" tabindex="-1"></a> <span class="st">"repetition_penalty"</span>: <span class="fl">1.03</span>,</span>
	<span id="cb19-14"><a href="#cb19-14" aria-hidden="true" tabindex="-1"></a> },</span>
	<span id="cb19-15"><a href="#cb19-15" aria-hidden="true" tabindex="-1"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="143d4d0b" class="cell" data-execution_count="20">
	<div class="sourceCode cell-code" id="cb20"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb20-1"><a href="#cb20-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> ragatouille <span class="im">import</span> RAGPretrainedModel</span>
	<span id="cb20-2"><a href="#cb20-2" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_core.vectorstores <span class="im">import</span> VectorStore</span>
	<span id="cb20-3"><a href="#cb20-3" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain_core.language_models.llms <span class="im">import</span> LLM</span>
	<span id="cb20-4"><a href="#cb20-4" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-5"><a href="#cb20-5" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-6"><a href="#cb20-6" aria-hidden="true" tabindex="-1"></a><span class="kw">def</span> answer_with_rag(</span>
	<span id="cb20-7"><a href="#cb20-7" aria-hidden="true" tabindex="-1"></a> question: <span class="bu">str</span>,</span>
	<span id="cb20-8"><a href="#cb20-8" aria-hidden="true" tabindex="-1"></a> llm: LLM,</span>
	<span id="cb20-9"><a href="#cb20-9" aria-hidden="true" tabindex="-1"></a> knowledge_index: VectorStore,</span>
	<span id="cb20-10"><a href="#cb20-10" aria-hidden="true" tabindex="-1"></a> reranker: Optional[RAGPretrainedModel] <span class="op">=</span> <span class="va">None</span>,</span>
	<span id="cb20-11"><a href="#cb20-11" aria-hidden="true" tabindex="-1"></a> num_retrieved_docs: <span class="bu">int</span> <span class="op">=</span> <span class="dv">30</span>,</span>
	<span id="cb20-12"><a href="#cb20-12" aria-hidden="true" tabindex="-1"></a> num_docs_final: <span class="bu">int</span> <span class="op">=</span> <span class="dv">7</span>,</span>
	<span id="cb20-13"><a href="#cb20-13" aria-hidden="true" tabindex="-1"></a>) <span class="op">-></span> Tuple[<span class="bu">str</span>, List[LangchainDocument]]:</span>
	<span id="cb20-14"><a href="#cb20-14" aria-hidden="true" tabindex="-1"></a> <span class="co">"""Answer a question using RAG with the given knowledge index."""</span></span>
	<span id="cb20-15"><a href="#cb20-15" aria-hidden="true" tabindex="-1"></a> <span class="co"># Gather documents with retriever</span></span>
	<span id="cb20-16"><a href="#cb20-16" aria-hidden="true" tabindex="-1"></a> relevant_docs <span class="op">=</span> knowledge_index.similarity_search(query<span class="op">=</span>question, k<span class="op">=</span>num_retrieved_docs)</span>
	<span id="cb20-17"><a href="#cb20-17" aria-hidden="true" tabindex="-1"></a> relevant_docs <span class="op">=</span> [doc.page_content <span class="cf">for</span> doc <span class="kw">in</span> relevant_docs] <span class="co"># keep only the text</span></span>
	<span id="cb20-18"><a href="#cb20-18" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-19"><a href="#cb20-19" aria-hidden="true" tabindex="-1"></a> <span class="co"># Optionally rerank results</span></span>
	<span id="cb20-20"><a href="#cb20-20" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> reranker:</span>
	<span id="cb20-21"><a href="#cb20-21" aria-hidden="true" tabindex="-1"></a> relevant_docs <span class="op">=</span> reranker.rerank(question, relevant_docs, k<span class="op">=</span>num_docs_final)</span>
	<span id="cb20-22"><a href="#cb20-22" aria-hidden="true" tabindex="-1"></a> relevant_docs <span class="op">=</span> [doc[<span class="st">"content"</span>] <span class="cf">for</span> doc <span class="kw">in</span> relevant_docs]</span>
	<span id="cb20-23"><a href="#cb20-23" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-24"><a href="#cb20-24" aria-hidden="true" tabindex="-1"></a> relevant_docs <span class="op">=</span> relevant_docs[:num_docs_final]</span>
	<span id="cb20-25"><a href="#cb20-25" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-26"><a href="#cb20-26" aria-hidden="true" tabindex="-1"></a> <span class="co"># Build the final prompt</span></span>
	<span id="cb20-27"><a href="#cb20-27" aria-hidden="true" tabindex="-1"></a> context <span class="op">=</span> <span class="st">"</span><span class="ch">\n</span><span class="st">Extracted documents:</span><span class="ch">\n</span><span class="st">"</span></span>
	<span id="cb20-28"><a href="#cb20-28" aria-hidden="true" tabindex="-1"></a> context <span class="op">+=</span> <span class="st">""</span>.join([<span class="ss">f"Document </span><span class="sc">{</span><span class="bu">str</span>(i)<span class="sc">}</span><span class="ss">:::</span><span class="ch">\n</span><span class="ss">"</span> <span class="op">+</span> doc <span class="cf">for</span> i, doc <span class="kw">in</span> <span class="bu">enumerate</span>(relevant_docs)])</span>
	<span id="cb20-29"><a href="#cb20-29" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-30"><a href="#cb20-30" aria-hidden="true" tabindex="-1"></a> final_prompt <span class="op">=</span> RAG_PROMPT_TEMPLATE.<span class="bu">format</span>(question<span class="op">=</span>question, context<span class="op">=</span>context)</span>
	<span id="cb20-31"><a href="#cb20-31" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-32"><a href="#cb20-32" aria-hidden="true" tabindex="-1"></a> <span class="co"># Redact an answer</span></span>
	<span id="cb20-33"><a href="#cb20-33" aria-hidden="true" tabindex="-1"></a> answer <span class="op">=</span> llm(final_prompt)</span>
	<span id="cb20-34"><a href="#cb20-34" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb20-35"><a href="#cb20-35" aria-hidden="true" tabindex="-1"></a> <span class="cf">return</span> answer, relevant_docs</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	</section>
	<section id="benchmarking-the-rag-system" class="level1">
	<h1>3. Benchmarking the RAG system</h1>
	<p>The RAG system and the evaluation datasets are now ready. The last step is to judge the RAG system’s output on this evlauation dataset.</p>
	<p>To this end, <strong>we setup a judge agent</strong>. ⚖️🤖</p>
	<p>Out of <a href="https://docs.ragas.io/en/latest/concepts/metrics/index.html">the different RAG evaluation metrics</a>, we choose to focus only on faithfulness since it the best end-to-end metric of our system’s performance.</p>
	<blockquote class="blockquote">
	<p>We use GPT4 as a judge for its empirically good performance, but you could try with other models such as <a href="https://huggingface.co/kaist-ai/prometheus-13b-v1.0">kaist-ai/prometheus-13b-v1.0</a> or <a href="https://huggingface.co/BAAI/JudgeLM-33B-v1.0">BAAI/JudgeLM-33B-v1.0</a>.</p>
	</blockquote>
	<p>💡 <em>In the evaluation prompt, we give a detailed description each metric on the scale 1-5, as is done in <a href="https://huggingface.co/kaist-ai/prometheus-13b-v1.0">Prometheus’s prompt template</a>: this helps the model ground its metric precisely. If instead you give the judge LLM a vague scale to work with, the outputs will not be consistent enough between different examples.</em></p>
	<p>💡 <em>Again, prompting the LLM to output rationale before giving its final score gives it more tokens to help it formalize and elaborate a judgement.</em></p>
	<div id="ce53b5ec" class="cell" data-execution_count="21">
	<div class="sourceCode cell-code" id="cb21"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb21-1"><a href="#cb21-1" aria-hidden="true" tabindex="-1"></a><span class="kw">def</span> run_rag_tests(</span>
	<span id="cb21-2"><a href="#cb21-2" aria-hidden="true" tabindex="-1"></a> eval_dataset: datasets.Dataset,</span>
	<span id="cb21-3"><a href="#cb21-3" aria-hidden="true" tabindex="-1"></a> llm: BaseChatModel,</span>
	<span id="cb21-4"><a href="#cb21-4" aria-hidden="true" tabindex="-1"></a> knowledge_index: VectorStore,</span>
	<span id="cb21-5"><a href="#cb21-5" aria-hidden="true" tabindex="-1"></a> output_file: <span class="bu">str</span>,</span>
	<span id="cb21-6"><a href="#cb21-6" aria-hidden="true" tabindex="-1"></a> reranker: Optional[RAGPretrainedModel] <span class="op">=</span> <span class="va">None</span>,</span>
	<span id="cb21-7"><a href="#cb21-7" aria-hidden="true" tabindex="-1"></a> verbose: Optional[<span class="bu">bool</span>] <span class="op">=</span> <span class="va">True</span>,</span>
	<span id="cb21-8"><a href="#cb21-8" aria-hidden="true" tabindex="-1"></a> test_settings: Optional[<span class="bu">str</span>] <span class="op">=</span> <span class="va">None</span>, <span class="co"># To document the test settings used</span></span>
	<span id="cb21-9"><a href="#cb21-9" aria-hidden="true" tabindex="-1"></a>):</span>
	<span id="cb21-10"><a href="#cb21-10" aria-hidden="true" tabindex="-1"></a> <span class="co">"""Runs RAG tests on the given dataset and saves the results to the given output file."""</span></span>
	<span id="cb21-11"><a href="#cb21-11" aria-hidden="true" tabindex="-1"></a> <span class="cf">try</span>: <span class="co"># load previous generations if they exist</span></span>
	<span id="cb21-12"><a href="#cb21-12" aria-hidden="true" tabindex="-1"></a> <span class="cf">with</span> <span class="bu">open</span>(output_file, <span class="st">"r"</span>) <span class="im">as</span> f:</span>
	<span id="cb21-13"><a href="#cb21-13" aria-hidden="true" tabindex="-1"></a> outputs <span class="op">=</span> json.load(f)</span>
	<span id="cb21-14"><a href="#cb21-14" aria-hidden="true" tabindex="-1"></a> <span class="cf">except</span>:</span>
	<span id="cb21-15"><a href="#cb21-15" aria-hidden="true" tabindex="-1"></a> outputs <span class="op">=</span> []</span>
	<span id="cb21-16"><a href="#cb21-16" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb21-17"><a href="#cb21-17" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> example <span class="kw">in</span> tqdm(eval_dataset):</span>
	<span id="cb21-18"><a href="#cb21-18" aria-hidden="true" tabindex="-1"></a> question <span class="op">=</span> example[<span class="st">"question"</span>]</span>
	<span id="cb21-19"><a href="#cb21-19" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> question <span class="kw">in</span> [output[<span class="st">"question"</span>] <span class="cf">for</span> output <span class="kw">in</span> outputs]:</span>
	<span id="cb21-20"><a href="#cb21-20" aria-hidden="true" tabindex="-1"></a> <span class="cf">continue</span></span>
	<span id="cb21-21"><a href="#cb21-21" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb21-22"><a href="#cb21-22" aria-hidden="true" tabindex="-1"></a> answer, relevant_docs <span class="op">=</span> answer_with_rag(question, llm, knowledge_index, reranker<span class="op">=</span>reranker)</span>
	<span id="cb21-23"><a href="#cb21-23" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> verbose:</span>
	<span id="cb21-24"><a href="#cb21-24" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="st">"======================================================="</span>)</span>
	<span id="cb21-25"><a href="#cb21-25" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="ss">f"Question: </span><span class="sc">{</span>question<span class="sc">}</span><span class="ss">"</span>)</span>
	<span id="cb21-26"><a href="#cb21-26" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="ss">f"Answer: </span><span class="sc">{</span>answer<span class="sc">}</span><span class="ss">"</span>)</span>
	<span id="cb21-27"><a href="#cb21-27" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="ss">f'True answer: </span><span class="sc">{</span>example[<span class="st">"answer"</span>]<span class="sc">}</span><span class="ss">'</span>)</span>
	<span id="cb21-28"><a href="#cb21-28" aria-hidden="true" tabindex="-1"></a> result <span class="op">=</span> {</span>
	<span id="cb21-29"><a href="#cb21-29" aria-hidden="true" tabindex="-1"></a> <span class="st">"question"</span>: question,</span>
	<span id="cb21-30"><a href="#cb21-30" aria-hidden="true" tabindex="-1"></a> <span class="st">"true_answer"</span>: example[<span class="st">"answer"</span>],</span>
	<span id="cb21-31"><a href="#cb21-31" aria-hidden="true" tabindex="-1"></a> <span class="st">"source_doc"</span>: example[<span class="st">"source_doc"</span>],</span>
	<span id="cb21-32"><a href="#cb21-32" aria-hidden="true" tabindex="-1"></a> <span class="st">"generated_answer"</span>: answer,</span>
	<span id="cb21-33"><a href="#cb21-33" aria-hidden="true" tabindex="-1"></a> <span class="st">"retrieved_docs"</span>: [doc <span class="cf">for</span> doc <span class="kw">in</span> relevant_docs],</span>
	<span id="cb21-34"><a href="#cb21-34" aria-hidden="true" tabindex="-1"></a> }</span>
	<span id="cb21-35"><a href="#cb21-35" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> test_settings:</span>
	<span id="cb21-36"><a href="#cb21-36" aria-hidden="true" tabindex="-1"></a> result[<span class="st">"test_settings"</span>] <span class="op">=</span> test_settings</span>
	<span id="cb21-37"><a href="#cb21-37" aria-hidden="true" tabindex="-1"></a> outputs.append(result)</span>
	<span id="cb21-38"><a href="#cb21-38" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb21-39"><a href="#cb21-39" aria-hidden="true" tabindex="-1"></a> <span class="cf">with</span> <span class="bu">open</span>(output_file, <span class="st">"w"</span>) <span class="im">as</span> f:</span>
	<span id="cb21-40"><a href="#cb21-40" aria-hidden="true" tabindex="-1"></a> json.dump(outputs, f)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="49f7ef44" class="cell" data-execution_count="22">
	<div class="sourceCode cell-code" id="cb22"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb22-1"><a href="#cb22-1" aria-hidden="true" tabindex="-1"></a>EVALUATION_PROMPT <span class="op">=</span> <span class="st">"""###Task Description:</span></span>
	<span id="cb22-2"><a href="#cb22-2" aria-hidden="true" tabindex="-1"></a><span class="st">An instruction (might include an Input inside it), a response to evaluate, a reference answer that gets a score of 5, and a score rubric representing a evaluation criteria are given.</span></span>
	<span id="cb22-3"><a href="#cb22-3" aria-hidden="true" tabindex="-1"></a><span class="st">1. Write a detailed feedback that assess the quality of the response strictly based on the given score rubric, not evaluating in general.</span></span>
	<span id="cb22-4"><a href="#cb22-4" aria-hidden="true" tabindex="-1"></a><span class="st">2. After writing a feedback, write a score that is an integer between 1 and 5. You should refer to the score rubric.</span></span>
	<span id="cb22-5"><a href="#cb22-5" aria-hidden="true" tabindex="-1"></a><span class="st">3. The output format should look as follows: </span><span class="ch">\"</span><span class="st">Feedback: </span><span class="sc">{{</span><span class="st">write a feedback for criteria</span><span class="sc">}}</span><span class="st"> [RESULT] </span><span class="sc">{{</span><span class="st">an integer number between 1 and 5</span><span class="sc">}}</span><span class="ch">\"</span></span>
	<span id="cb22-6"><a href="#cb22-6" aria-hidden="true" tabindex="-1"></a><span class="st">4. Please do not generate any other opening, closing, and explanations. Be sure to include [RESULT] in your output.</span></span>
	<span id="cb22-7"><a href="#cb22-7" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-8"><a href="#cb22-8" aria-hidden="true" tabindex="-1"></a><span class="st">###The instruction to evaluate:</span></span>
	<span id="cb22-9"><a href="#cb22-9" aria-hidden="true" tabindex="-1"></a><span class="sc">{instruction}</span></span>
	<span id="cb22-10"><a href="#cb22-10" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-11"><a href="#cb22-11" aria-hidden="true" tabindex="-1"></a><span class="st">###Response to evaluate:</span></span>
	<span id="cb22-12"><a href="#cb22-12" aria-hidden="true" tabindex="-1"></a><span class="sc">{response}</span></span>
	<span id="cb22-13"><a href="#cb22-13" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-14"><a href="#cb22-14" aria-hidden="true" tabindex="-1"></a><span class="st">###Reference Answer (Score 5):</span></span>
	<span id="cb22-15"><a href="#cb22-15" aria-hidden="true" tabindex="-1"></a><span class="sc">{reference_answer}</span></span>
	<span id="cb22-16"><a href="#cb22-16" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-17"><a href="#cb22-17" aria-hidden="true" tabindex="-1"></a><span class="st">###Score Rubrics:</span></span>
	<span id="cb22-18"><a href="#cb22-18" aria-hidden="true" tabindex="-1"></a><span class="st">[Is the response correct, accurate, and factual based on the reference answer?]</span></span>
	<span id="cb22-19"><a href="#cb22-19" aria-hidden="true" tabindex="-1"></a><span class="st">Score 1: The response is completely incorrect, inaccurate, and/or not factual.</span></span>
	<span id="cb22-20"><a href="#cb22-20" aria-hidden="true" tabindex="-1"></a><span class="st">Score 2: The response is mostly incorrect, inaccurate, and/or not factual.</span></span>
	<span id="cb22-21"><a href="#cb22-21" aria-hidden="true" tabindex="-1"></a><span class="st">Score 3: The response is somewhat correct, accurate, and/or factual.</span></span>
	<span id="cb22-22"><a href="#cb22-22" aria-hidden="true" tabindex="-1"></a><span class="st">Score 4: The response is mostly correct, accurate, and factual.</span></span>
	<span id="cb22-23"><a href="#cb22-23" aria-hidden="true" tabindex="-1"></a><span class="st">Score 5: The response is completely correct, accurate, and factual.</span></span>
	<span id="cb22-24"><a href="#cb22-24" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-25"><a href="#cb22-25" aria-hidden="true" tabindex="-1"></a><span class="st">###Feedback:"""</span></span>
	<span id="cb22-26"><a href="#cb22-26" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-27"><a href="#cb22-27" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.prompts.chat <span class="im">import</span> (</span>
	<span id="cb22-28"><a href="#cb22-28" aria-hidden="true" tabindex="-1"></a> ChatPromptTemplate,</span>
	<span id="cb22-29"><a href="#cb22-29" aria-hidden="true" tabindex="-1"></a> HumanMessagePromptTemplate,</span>
	<span id="cb22-30"><a href="#cb22-30" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb22-31"><a href="#cb22-31" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.schema <span class="im">import</span> SystemMessage</span>
	<span id="cb22-32"><a href="#cb22-32" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-33"><a href="#cb22-33" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb22-34"><a href="#cb22-34" aria-hidden="true" tabindex="-1"></a>evaluation_prompt_template <span class="op">=</span> ChatPromptTemplate.from_messages(</span>
	<span id="cb22-35"><a href="#cb22-35" aria-hidden="true" tabindex="-1"></a> [</span>
	<span id="cb22-36"><a href="#cb22-36" aria-hidden="true" tabindex="-1"></a> SystemMessage(content<span class="op">=</span><span class="st">"You are a fair evaluator language model."</span>),</span>
	<span id="cb22-37"><a href="#cb22-37" aria-hidden="true" tabindex="-1"></a> HumanMessagePromptTemplate.from_template(EVALUATION_PROMPT),</span>
	<span id="cb22-38"><a href="#cb22-38" aria-hidden="true" tabindex="-1"></a> ]</span>
	<span id="cb22-39"><a href="#cb22-39" aria-hidden="true" tabindex="-1"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="b678d8fe" class="cell" data-execution_count="23">
	<div class="sourceCode cell-code" id="cb23"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb23-1"><a href="#cb23-1" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> langchain.chat_models <span class="im">import</span> ChatOpenAI</span>
	<span id="cb23-2"><a href="#cb23-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-3"><a href="#cb23-3" aria-hidden="true" tabindex="-1"></a>eval_chat_model <span class="op">=</span> ChatOpenAI(model<span class="op">=</span><span class="st">"gpt-4-1106-preview"</span>, temperature<span class="op">=</span><span class="dv">0</span>)</span>
	<span id="cb23-4"><a href="#cb23-4" aria-hidden="true" tabindex="-1"></a>evaluator_name <span class="op">=</span> <span class="st">"GPT4"</span></span>
	<span id="cb23-5"><a href="#cb23-5" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-6"><a href="#cb23-6" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-7"><a href="#cb23-7" aria-hidden="true" tabindex="-1"></a><span class="kw">def</span> evaluate_answers(</span>
	<span id="cb23-8"><a href="#cb23-8" aria-hidden="true" tabindex="-1"></a> answer_path: <span class="bu">str</span>,</span>
	<span id="cb23-9"><a href="#cb23-9" aria-hidden="true" tabindex="-1"></a> eval_chat_model: BaseChatModel,</span>
	<span id="cb23-10"><a href="#cb23-10" aria-hidden="true" tabindex="-1"></a> evaluator_name: <span class="bu">str</span>,</span>
	<span id="cb23-11"><a href="#cb23-11" aria-hidden="true" tabindex="-1"></a> evaluation_prompt_template: ChatPromptTemplate,</span>
	<span id="cb23-12"><a href="#cb23-12" aria-hidden="true" tabindex="-1"></a>) <span class="op">-></span> <span class="va">None</span>:</span>
	<span id="cb23-13"><a href="#cb23-13" aria-hidden="true" tabindex="-1"></a> <span class="co">"""Evaluates generated answers. Modifies the given answer file in place for better checkpointing."""</span></span>
	<span id="cb23-14"><a href="#cb23-14" aria-hidden="true" tabindex="-1"></a> answers <span class="op">=</span> []</span>
	<span id="cb23-15"><a href="#cb23-15" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> os.path.isfile(answer_path): <span class="co"># load previous generations if they exist</span></span>
	<span id="cb23-16"><a href="#cb23-16" aria-hidden="true" tabindex="-1"></a> answers <span class="op">=</span> json.load(<span class="bu">open</span>(answer_path, <span class="st">"r"</span>))</span>
	<span id="cb23-17"><a href="#cb23-17" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-18"><a href="#cb23-18" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> experiment <span class="kw">in</span> tqdm(answers):</span>
	<span id="cb23-19"><a href="#cb23-19" aria-hidden="true" tabindex="-1"></a> <span class="cf">if</span> <span class="ss">f"eval_score_</span><span class="sc">{</span>evaluator_name<span class="sc">}</span><span class="ss">"</span> <span class="kw">in</span> experiment:</span>
	<span id="cb23-20"><a href="#cb23-20" aria-hidden="true" tabindex="-1"></a> <span class="cf">continue</span></span>
	<span id="cb23-21"><a href="#cb23-21" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-22"><a href="#cb23-22" aria-hidden="true" tabindex="-1"></a> eval_prompt <span class="op">=</span> evaluation_prompt_template.format_messages(</span>
	<span id="cb23-23"><a href="#cb23-23" aria-hidden="true" tabindex="-1"></a> instruction<span class="op">=</span>experiment[<span class="st">"question"</span>],</span>
	<span id="cb23-24"><a href="#cb23-24" aria-hidden="true" tabindex="-1"></a> response<span class="op">=</span>experiment[<span class="st">"generated_answer"</span>],</span>
	<span id="cb23-25"><a href="#cb23-25" aria-hidden="true" tabindex="-1"></a> reference_answer<span class="op">=</span>experiment[<span class="st">"true_answer"</span>],</span>
	<span id="cb23-26"><a href="#cb23-26" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb23-27"><a href="#cb23-27" aria-hidden="true" tabindex="-1"></a> eval_result <span class="op">=</span> eval_chat_model.invoke(eval_prompt)</span>
	<span id="cb23-28"><a href="#cb23-28" aria-hidden="true" tabindex="-1"></a> feedback, score <span class="op">=</span> [item.strip() <span class="cf">for</span> item <span class="kw">in</span> eval_result.content.split(<span class="st">"[RESULT]"</span>)]</span>
	<span id="cb23-29"><a href="#cb23-29" aria-hidden="true" tabindex="-1"></a> experiment[<span class="ss">f"eval_score_</span><span class="sc">{</span>evaluator_name<span class="sc">}</span><span class="ss">"</span>] <span class="op">=</span> score</span>
	<span id="cb23-30"><a href="#cb23-30" aria-hidden="true" tabindex="-1"></a> experiment[<span class="ss">f"eval_feedback_</span><span class="sc">{</span>evaluator_name<span class="sc">}</span><span class="ss">"</span>] <span class="op">=</span> feedback</span>
	<span id="cb23-31"><a href="#cb23-31" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb23-32"><a href="#cb23-32" aria-hidden="true" tabindex="-1"></a> <span class="cf">with</span> <span class="bu">open</span>(answer_path, <span class="st">"w"</span>) <span class="im">as</span> f:</span>
	<span id="cb23-33"><a href="#cb23-33" aria-hidden="true" tabindex="-1"></a> json.dump(answers, f)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<p>🚀 Let’s run the tests and evaluate answers!👇</p>
	<div id="55f9f502" class="cell" data-execution_count="24">
	<div class="sourceCode cell-code" id="cb24"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb24-1"><a href="#cb24-1" aria-hidden="true" tabindex="-1"></a><span class="cf">if</span> <span class="kw">not</span> os.path.exists(<span class="st">"./output"</span>):</span>
	<span id="cb24-2"><a href="#cb24-2" aria-hidden="true" tabindex="-1"></a> os.mkdir(<span class="st">"./output"</span>)</span>
	<span id="cb24-3"><a href="#cb24-3" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb24-4"><a href="#cb24-4" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> chunk_size <span class="kw">in</span> [<span class="dv">200</span>]: <span class="co"># Add other chunk sizes (in tokens) as needed</span></span>
	<span id="cb24-5"><a href="#cb24-5" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> embeddings <span class="kw">in</span> [<span class="st">"thenlper/gte-small"</span>]: <span class="co"># Add other embeddings as needed</span></span>
	<span id="cb24-6"><a href="#cb24-6" aria-hidden="true" tabindex="-1"></a> <span class="cf">for</span> rerank <span class="kw">in</span> [<span class="va">True</span>, <span class="va">False</span>]:</span>
	<span id="cb24-7"><a href="#cb24-7" aria-hidden="true" tabindex="-1"></a> settings_name <span class="op">=</span> <span class="ss">f"chunk:</span><span class="sc">{</span>chunk_size<span class="sc">}</span><span class="ss">_embeddings:</span><span class="sc">{</span>embeddings<span class="sc">.</span>replace(<span class="st">'/'</span>, <span class="st">'~'</span>)<span class="sc">}</span><span class="ss">_rerank:</span><span class="sc">{</span>rerank<span class="sc">}</span><span class="ss">_reader-model:</span><span class="sc">{</span>READER_MODEL_NAME<span class="sc">}</span><span class="ss">"</span></span>
	<span id="cb24-8"><a href="#cb24-8" aria-hidden="true" tabindex="-1"></a> output_file_name <span class="op">=</span> <span class="ss">f"./output/rag_</span><span class="sc">{</span>settings_name<span class="sc">}</span><span class="ss">.json"</span></span>
	<span id="cb24-9"><a href="#cb24-9" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb24-10"><a href="#cb24-10" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="ss">f"Running evaluation for </span><span class="sc">{</span>settings_name<span class="sc">}</span><span class="ss">:"</span>)</span>
	<span id="cb24-11"><a href="#cb24-11" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb24-12"><a href="#cb24-12" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="st">"Loading knowledge base embeddings..."</span>)</span>
	<span id="cb24-13"><a href="#cb24-13" aria-hidden="true" tabindex="-1"></a> knowledge_index <span class="op">=</span> load_embeddings(</span>
	<span id="cb24-14"><a href="#cb24-14" aria-hidden="true" tabindex="-1"></a> RAW_KNOWLEDGE_BASE,</span>
	<span id="cb24-15"><a href="#cb24-15" aria-hidden="true" tabindex="-1"></a> chunk_size<span class="op">=</span>chunk_size,</span>
	<span id="cb24-16"><a href="#cb24-16" aria-hidden="true" tabindex="-1"></a> embedding_model_name<span class="op">=</span>embeddings,</span>
	<span id="cb24-17"><a href="#cb24-17" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb24-18"><a href="#cb24-18" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb24-19"><a href="#cb24-19" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="st">"Running RAG..."</span>)</span>
	<span id="cb24-20"><a href="#cb24-20" aria-hidden="true" tabindex="-1"></a> reranker <span class="op">=</span> (</span>
	<span id="cb24-21"><a href="#cb24-21" aria-hidden="true" tabindex="-1"></a> RAGPretrainedModel.from_pretrained(<span class="st">"colbert-ir/colbertv2.0"</span>) <span class="cf">if</span> rerank <span class="cf">else</span> <span class="va">None</span></span>
	<span id="cb24-22"><a href="#cb24-22" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb24-23"><a href="#cb24-23" aria-hidden="true" tabindex="-1"></a> run_rag_tests(</span>
	<span id="cb24-24"><a href="#cb24-24" aria-hidden="true" tabindex="-1"></a> eval_dataset<span class="op">=</span>eval_dataset,</span>
	<span id="cb24-25"><a href="#cb24-25" aria-hidden="true" tabindex="-1"></a> llm<span class="op">=</span>READER_LLM,</span>
	<span id="cb24-26"><a href="#cb24-26" aria-hidden="true" tabindex="-1"></a> knowledge_index<span class="op">=</span>knowledge_index,</span>
	<span id="cb24-27"><a href="#cb24-27" aria-hidden="true" tabindex="-1"></a> output_file<span class="op">=</span>output_file_name,</span>
	<span id="cb24-28"><a href="#cb24-28" aria-hidden="true" tabindex="-1"></a> reranker<span class="op">=</span>reranker,</span>
	<span id="cb24-29"><a href="#cb24-29" aria-hidden="true" tabindex="-1"></a> verbose<span class="op">=</span><span class="va">False</span>,</span>
	<span id="cb24-30"><a href="#cb24-30" aria-hidden="true" tabindex="-1"></a> test_settings<span class="op">=</span>settings_name,</span>
	<span id="cb24-31"><a href="#cb24-31" aria-hidden="true" tabindex="-1"></a> )</span>
	<span id="cb24-32"><a href="#cb24-32" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb24-33"><a href="#cb24-33" aria-hidden="true" tabindex="-1"></a> <span class="bu">print</span>(<span class="st">"Running evaluation..."</span>)</span>
	<span id="cb24-34"><a href="#cb24-34" aria-hidden="true" tabindex="-1"></a> evaluate_answers(</span>
	<span id="cb24-35"><a href="#cb24-35" aria-hidden="true" tabindex="-1"></a> output_file_name,</span>
	<span id="cb24-36"><a href="#cb24-36" aria-hidden="true" tabindex="-1"></a> eval_chat_model,</span>
	<span id="cb24-37"><a href="#cb24-37" aria-hidden="true" tabindex="-1"></a> evaluator_name,</span>
	<span id="cb24-38"><a href="#cb24-38" aria-hidden="true" tabindex="-1"></a> evaluation_prompt_template,</span>
	<span id="cb24-39"><a href="#cb24-39" aria-hidden="true" tabindex="-1"></a> )</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<section id="inspect-results" class="level3">
	<h3 class="anchored" data-anchor-id="inspect-results">Inspect results</h3>
	<div id="9fbbe7e3" class="cell" data-execution_count="25">
	<div class="sourceCode cell-code" id="cb25"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb25-1"><a href="#cb25-1" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> glob</span>
	<span id="cb25-2"><a href="#cb25-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb25-3"><a href="#cb25-3" aria-hidden="true" tabindex="-1"></a>outputs <span class="op">=</span> []</span>
	<span id="cb25-4"><a href="#cb25-4" aria-hidden="true" tabindex="-1"></a><span class="cf">for</span> <span class="bu">file</span> <span class="kw">in</span> glob.glob(<span class="st">"./output/*.json"</span>):</span>
	<span id="cb25-5"><a href="#cb25-5" aria-hidden="true" tabindex="-1"></a> output <span class="op">=</span> pd.DataFrame(json.load(<span class="bu">open</span>(<span class="bu">file</span>, <span class="st">"r"</span>)))</span>
	<span id="cb25-6"><a href="#cb25-6" aria-hidden="true" tabindex="-1"></a> output[<span class="st">"settings"</span>] <span class="op">=</span> <span class="bu">file</span></span>
	<span id="cb25-7"><a href="#cb25-7" aria-hidden="true" tabindex="-1"></a> outputs.append(output)</span>
	<span id="cb25-8"><a href="#cb25-8" aria-hidden="true" tabindex="-1"></a>result <span class="op">=</span> pd.concat(outputs)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="8e32ac07" class="cell" data-execution_count="26">
	<div class="sourceCode cell-code" id="cb26"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb26-1"><a href="#cb26-1" aria-hidden="true" tabindex="-1"></a>result[<span class="st">"eval_score_GPT4"</span>] <span class="op">=</span> result[<span class="st">"eval_score_GPT4"</span>].<span class="bu">apply</span>(</span>
	<span id="cb26-2"><a href="#cb26-2" aria-hidden="true" tabindex="-1"></a> <span class="kw">lambda</span> x: <span class="bu">int</span>(x) <span class="cf">if</span> <span class="bu">isinstance</span>(x, <span class="bu">str</span>) <span class="cf">else</span> <span class="dv">1</span></span>
	<span id="cb26-3"><a href="#cb26-3" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb26-4"><a href="#cb26-4" aria-hidden="true" tabindex="-1"></a>result[<span class="st">"eval_score_GPT4"</span>] <span class="op">=</span> (result[<span class="st">"eval_score_GPT4"</span>] <span class="op">-</span> <span class="dv">1</span>) <span class="op">/</span> <span class="dv">4</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="14c33cb4" class="cell" data-execution_count="27">
	<div class="sourceCode cell-code" id="cb27"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb27-1"><a href="#cb27-1" aria-hidden="true" tabindex="-1"></a>average_scores <span class="op">=</span> result.groupby(<span class="st">"settings"</span>)[<span class="st">"eval_score_GPT4"</span>].mean()</span>
	<span id="cb27-2"><a href="#cb27-2" aria-hidden="true" tabindex="-1"></a>average_scores.sort_values()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	</section>
	<section id="example-results" class="level2">
	<h2 class="anchored" data-anchor-id="example-results">Example results</h2>
	<p>Let us load the results that I obtained by tweaking the different options available in this notebook. For more detail on why these options could work on not, see the notebook on <a href="advanced_rag">advanced_RAG</a>.</p>
	<p>As you can see in the graph below, some tweaks do not bring any improvement, some give huge performance boosts.</p>
	<p>➡️ <strong><em>There is no single good recipe: you should try several different directions when tuning your RAG systems.</em></strong></p>
	<div id="0f27b105" class="cell" data-execution_count="28">
	<div class="sourceCode cell-code" id="cb28"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb28-1"><a href="#cb28-1" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> plotly.express <span class="im">as</span> px</span>
	<span id="cb28-2"><a href="#cb28-2" aria-hidden="true" tabindex="-1"></a></span>
	<span id="cb28-3"><a href="#cb28-3" aria-hidden="true" tabindex="-1"></a>scores <span class="op">=</span> datasets.load_dataset(<span class="st">"m-ric/rag_scores_cookbook"</span>, split<span class="op">=</span><span class="st">"train"</span>)</span>
	<span id="cb28-4"><a href="#cb28-4" aria-hidden="true" tabindex="-1"></a>scores <span class="op">=</span> pd.Series(scores[<span class="st">"score"</span>], index<span class="op">=</span>scores[<span class="st">"settings"</span>])</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<div id="abdb166c" class="cell" data-execution_count="29">
	<div class="sourceCode cell-code" id="cb29"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb29-1"><a href="#cb29-1" aria-hidden="true" tabindex="-1"></a>fig <span class="op">=</span> px.bar(</span>
	<span id="cb29-2"><a href="#cb29-2" aria-hidden="true" tabindex="-1"></a> scores,</span>
	<span id="cb29-3"><a href="#cb29-3" aria-hidden="true" tabindex="-1"></a> color<span class="op">=</span>scores,</span>
	<span id="cb29-4"><a href="#cb29-4" aria-hidden="true" tabindex="-1"></a> labels<span class="op">=</span>{</span>
	<span id="cb29-5"><a href="#cb29-5" aria-hidden="true" tabindex="-1"></a> <span class="st">"value"</span>: <span class="st">"Accuracy"</span>,</span>
	<span id="cb29-6"><a href="#cb29-6" aria-hidden="true" tabindex="-1"></a> <span class="st">"settings"</span>: <span class="st">"Configuration"</span>,</span>
	<span id="cb29-7"><a href="#cb29-7" aria-hidden="true" tabindex="-1"></a> },</span>
	<span id="cb29-8"><a href="#cb29-8" aria-hidden="true" tabindex="-1"></a> color_continuous_scale<span class="op">=</span><span class="st">"bluered"</span>,</span>
	<span id="cb29-9"><a href="#cb29-9" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb29-10"><a href="#cb29-10" aria-hidden="true" tabindex="-1"></a>fig.update_layout(w</span>
	<span id="cb29-11"><a href="#cb29-11" aria-hidden="true" tabindex="-1"></a> width<span class="op">=</span><span class="dv">1000</span>,</span>
	<span id="cb29-12"><a href="#cb29-12" aria-hidden="true" tabindex="-1"></a> height<span class="op">=</span><span class="dv">600</span>,</span>
	<span id="cb29-13"><a href="#cb29-13" aria-hidden="true" tabindex="-1"></a> barmode<span class="op">=</span><span class="st">"group"</span>,</span>
	<span id="cb29-14"><a href="#cb29-14" aria-hidden="true" tabindex="-1"></a> yaxis_range<span class="op">=</span>[<span class="dv">0</span>, <span class="dv">100</span>],</span>
	<span id="cb29-15"><a href="#cb29-15" aria-hidden="true" tabindex="-1"></a> title<span class="op">=</span><span class="st">"<b>Accuracy of different RAG configurations</b>"</span>,</span>
	<span id="cb29-16"><a href="#cb29-16" aria-hidden="true" tabindex="-1"></a> xaxis_title<span class="op">=</span><span class="st">"RAG settings"</span>,</span>
	<span id="cb29-17"><a href="#cb29-17" aria-hidden="true" tabindex="-1"></a> font<span class="op">=</span><span class="bu">dict</span>(size<span class="op">=</span><span class="dv">15</span>),</span>
	<span id="cb29-18"><a href="#cb29-18" aria-hidden="true" tabindex="-1"></a>)</span>
	<span id="cb29-19"><a href="#cb29-19" aria-hidden="true" tabindex="-1"></a>fig.layout.yaxis.ticksuffix <span class="op">=</span> <span class="st">"%"</span></span>
	<span id="cb29-20"><a href="#cb29-20" aria-hidden="true" tabindex="-1"></a>fig.update_coloraxes(showscale<span class="op">=</span><span class="va">False</span>)</span>
	<span id="cb29-21"><a href="#cb29-21" aria-hidden="true" tabindex="-1"></a>fig.update_traces(texttemplate<span class="op">=</span><span class="st">"%</span><span class="sc">{y:.1f}</span><span class="st">"</span>, textposition<span class="op">=</span><span class="st">"outside"</span>)</span>
	<span id="cb29-22"><a href="#cb29-22" aria-hidden="true" tabindex="-1"></a>fig.show()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
	</div>
	<p><img src="https://huggingface.co/datasets/huggingface/cookbook-images/resolve/main/RAG_settings_accuracy.png" height="500" width="800"></p>
	<p>As you can see, these had varying impact on performance. In particular, tuning the chunk size is both easy and very impactful.</p>
	<p>But this is our case: your results could be very different: now that you have a robust evaluation pipeline, you can set on to explore other options! 🗺️</p>


	</section>
	</section>

	</main> <!-- /main -->
	<script id="quarto-html-after-body" type="application/javascript">
	window.document.addEventListener("DOMContentLoaded", function (event) {
	const toggleBodyColorMode = (bsSheetEl) => {
	const mode = bsSheetEl.getAttribute("data-mode");
	const bodyEl = window.document.querySelector("body");
	if (mode === "dark") {
	bodyEl.classList.add("quarto-dark");
	bodyEl.classList.remove("quarto-light");
	} else {
	bodyEl.classList.add("quarto-light");
	bodyEl.classList.remove("quarto-dark");
	}
	}
	const toggleBodyColorPrimary = () => {
	const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
	if (bsSheetEl) {
	toggleBodyColorMode(bsSheetEl);
	}
	}
	toggleBodyColorPrimary();
	const icon = "";
	const anchorJS = new window.AnchorJS();
	anchorJS.options = {
	placement: 'right',
	icon: icon
	};
	anchorJS.add('.anchored');
	const isCodeAnnotation = (el) => {
	for (const clz of el.classList) {
	if (clz.startsWith('code-annotation-')) {
	return true;
	}
	}
	return false;
	}
	const onCopySuccess = function(e) {
	// button target
	const button = e.trigger;
	// don't keep focus
	button.blur();
	// flash "checked"
	button.classList.add('code-copy-button-checked');
	var currentTitle = button.getAttribute("title");
	button.setAttribute("title", "Copied!");
	let tooltip;
	if (window.bootstrap) {
	button.setAttribute("data-bs-toggle", "tooltip");
	button.setAttribute("data-bs-placement", "left");
	button.setAttribute("data-bs-title", "Copied!");
	tooltip = new bootstrap.Tooltip(button,
	{ trigger: "manual",
	customClass: "code-copy-button-tooltip",
	offset: [0, -8]});
	tooltip.show();
	}
	setTimeout(function() {
	if (tooltip) {
	tooltip.hide();
	button.removeAttribute("data-bs-title");
	button.removeAttribute("data-bs-toggle");
	button.removeAttribute("data-bs-placement");
	}
	button.setAttribute("title", currentTitle);
	button.classList.remove('code-copy-button-checked');
	}, 1000);
	// clear code selection
	e.clearSelection();
	}
	const getTextToCopy = function(trigger) {
	const codeEl = trigger.previousElementSibling.cloneNode(true);
	for (const childEl of codeEl.children) {
	if (isCodeAnnotation(childEl)) {
	childEl.remove();
	}
	}
	return codeEl.innerText;
	}
	const clipboard = new window.ClipboardJS('.code-copy-button:not([data-in-quarto-modal])', {
	text: getTextToCopy
	});
	clipboard.on('success', onCopySuccess);
	if (window.document.getElementById('quarto-embedded-source-code-modal')) {
	const clipboardModal = new window.ClipboardJS('.code-copy-button[data-in-quarto-modal]', {
	text: getTextToCopy,
	container: window.document.getElementById('quarto-embedded-source-code-modal')
	});
	clipboardModal.on('success', onCopySuccess);
	}
	var localhostRegex = new RegExp(/^(?:http\|https):\/\/localhost\:?[0-9]*\//);
	var mailtoRegex = new RegExp(/^mailto:/);
	var filterRegex = new RegExp('/' + window.location.host + '/');
	var isInternal = (href) => {
	return filterRegex.test(href) \|\| localhostRegex.test(href) \|\| mailtoRegex.test(href);
	}
	// Inspect non-navigation links and adorn them if external
	var links = window.document.querySelectorAll('a[href]:not(.nav-link):not(.navbar-brand):not(.toc-action):not(.sidebar-link):not(.sidebar-item-toggle):not(.pagination-link):not(.no-external):not([aria-hidden]):not(.dropdown-item):not(.quarto-navigation-tool):not(.about-link)');
	for (var i=0; i<links.length; i++) {
	const link = links[i];
	if (!isInternal(link.href)) {
	// undo the damage that might have been done by quarto-nav.js in the case of
	// links that we want to consider external
	if (link.dataset.originalHref !== undefined) {
	link.href = link.dataset.originalHref;
	}
	}
	}
	function tippyHover(el, contentFn, onTriggerFn, onUntriggerFn) {
	const config = {
	allowHTML: true,
	maxWidth: 500,
	delay: 100,
	arrow: false,
	appendTo: function(el) {
	return el.parentElement;
	},
	interactive: true,
	interactiveBorder: 10,
	theme: 'quarto',
	placement: 'bottom-start',
	};
	if (contentFn) {
	config.content = contentFn;
	}
	if (onTriggerFn) {
	config.onTrigger = onTriggerFn;
	}
	if (onUntriggerFn) {
	config.onUntrigger = onUntriggerFn;
	}
	window.tippy(el, config);
	}
	const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
	for (var i=0; i<noterefs.length; i++) {
	const ref = noterefs[i];
	tippyHover(ref, function() {
	// use id or data attribute instead here
	let href = ref.getAttribute('data-footnote-href') \|\| ref.getAttribute('href');
	try { href = new URL(href).hash; } catch {}
	const id = href.replace(/^#\/?/, "");
	const note = window.document.getElementById(id);
	if (note) {
	return note.innerHTML;
	} else {
	return "";
	}
	});
	}
	const xrefs = window.document.querySelectorAll('a.quarto-xref');
	const processXRef = (id, note) => {
	// Strip column container classes
	const stripColumnClz = (el) => {
	el.classList.remove("page-full", "page-columns");
	if (el.children) {
	for (const child of el.children) {
	stripColumnClz(child);
	}
	}
	}
	stripColumnClz(note)
	if (id === null \|\| id.startsWith('sec-')) {
	// Special case sections, only their first couple elements
	const container = document.createElement("div");
	if (note.children && note.children.length > 2) {
	container.appendChild(note.children[0].cloneNode(true));
	for (let i = 1; i < note.children.length; i++) {
	const child = note.children[i];
	if (child.tagName === "P" && child.innerText === "") {
	continue;
	} else {
	container.appendChild(child.cloneNode(true));
	break;
	}
	}
	if (window.Quarto?.typesetMath) {
	window.Quarto.typesetMath(container);
	}
	return container.innerHTML
	} else {
	if (window.Quarto?.typesetMath) {
	window.Quarto.typesetMath(note);
	}
	return note.innerHTML;
	}
	} else {
	// Remove any anchor links if they are present
	const anchorLink = note.querySelector('a.anchorjs-link');
	if (anchorLink) {
	anchorLink.remove();
	}
	if (window.Quarto?.typesetMath) {
	window.Quarto.typesetMath(note);
	}
	if (note.classList.contains("callout")) {
	return note.outerHTML;
	} else {
	return note.innerHTML;
	}
	}
	}
	for (var i=0; i<xrefs.length; i++) {
	const xref = xrefs[i];
	tippyHover(xref, undefined, function(instance) {
	instance.disable();
	let url = xref.getAttribute('href');
	let hash = undefined;
	if (url.startsWith('#')) {
	hash = url;
	} else {
	try { hash = new URL(url).hash; } catch {}
	}
	if (hash) {
	const id = hash.replace(/^#\/?/, "");
	const note = window.document.getElementById(id);
	if (note !== null) {
	try {
	const html = processXRef(id, note.cloneNode(true));
	instance.setContent(html);
	} finally {
	instance.enable();
	instance.show();
	}
	} else {
	// See if we can fetch this
	fetch(url.split('#')[0])
	.then(res => res.text())
	.then(html => {
	const parser = new DOMParser();
	const htmlDoc = parser.parseFromString(html, "text/html");
	const note = htmlDoc.getElementById(id);
	if (note !== null) {
	const html = processXRef(id, note);
	instance.setContent(html);
	}
	}).finally(() => {
	instance.enable();
	instance.show();
	});
	}
	} else {
	// See if we can fetch a full url (with no hash to target)
	// This is a special case and we should probably do some content thinning / targeting
	fetch(url)
	.then(res => res.text())
	.then(html => {
	const parser = new DOMParser();
	const htmlDoc = parser.parseFromString(html, "text/html");
	const note = htmlDoc.querySelector('main.content');
	if (note !== null) {
	// This should only happen for chapter cross references
	// (since there is no id in the URL)
	// remove the first header
	if (note.children.length > 0 && note.children[0].tagName === "HEADER") {
	note.children[0].remove();
	}
	const html = processXRef(null, note);
	instance.setContent(html);
	}
	}).finally(() => {
	instance.enable();
	instance.show();
	});
	}
	}, function(instance) {
	});
	}
	let selectedAnnoteEl;
	const selectorForAnnotation = ( cell, annotation) => {
	let cellAttr = 'data-code-cell="' + cell + '"';
	let lineAttr = 'data-code-annotation="' + annotation + '"';
	const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
	return selector;
	}
	const selectCodeLines = (annoteEl) => {
	const doc = window.document;
	const targetCell = annoteEl.getAttribute("data-target-cell");
	const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
	const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
	const lines = annoteSpan.getAttribute("data-code-lines").split(",");
	const lineIds = lines.map((line) => {
	return targetCell + "-" + line;
	})
	let top = null;
	let height = null;
	let parent = null;
	if (lineIds.length > 0) {
	//compute the position of the single el (top and bottom and make a div)
	const el = window.document.getElementById(lineIds[0]);
	top = el.offsetTop;
	height = el.offsetHeight;
	parent = el.parentElement.parentElement;
	if (lineIds.length > 1) {
	const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
	const bottom = lastEl.offsetTop + lastEl.offsetHeight;
	height = bottom - top;
	}
	if (top !== null && height !== null && parent !== null) {
	// cook up a div (if necessary) and position it
	let div = window.document.getElementById("code-annotation-line-highlight");
	if (div === null) {
	div = window.document.createElement("div");
	div.setAttribute("id", "code-annotation-line-highlight");
	div.style.position = 'absolute';
	parent.appendChild(div);
	}
	div.style.top = top - 2 + "px";
	div.style.height = height + 4 + "px";
	div.style.left = 0;
	let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
	if (gutterDiv === null) {
	gutterDiv = window.document.createElement("div");
	gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
	gutterDiv.style.position = 'absolute';
	const codeCell = window.document.getElementById(targetCell);
	const gutter = codeCell.querySelector('.code-annotation-gutter');
	gutter.appendChild(gutterDiv);
	}
	gutterDiv.style.top = top - 2 + "px";
	gutterDiv.style.height = height + 4 + "px";
	}
	selectedAnnoteEl = annoteEl;
	}
	};
	const unselectCodeLines = () => {
	const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
	elementsIds.forEach((elId) => {
	const div = window.document.getElementById(elId);
	if (div) {
	div.remove();
	}
	});
	selectedAnnoteEl = undefined;
	};
	// Handle positioning of the toggle
	window.addEventListener(
	"resize",
	throttle(() => {
	elRect = undefined;
	if (selectedAnnoteEl) {
	selectCodeLines(selectedAnnoteEl);
	}
	}, 10)
	);
	function throttle(fn, ms) {
	let throttle = false;
	let timer;
	return (...args) => {
	if(!throttle) { // first call gets through
	fn.apply(this, args);
	throttle = true;
	} else { // all the others get throttled
	if(timer) clearTimeout(timer); // cancel #2
	timer = setTimeout(() => {
	fn.apply(this, args);
	timer = throttle = false;
	}, ms);
	}
	};
	}
	// Attach click handler to the DT
	const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
	for (const annoteDlNode of annoteDls) {
	annoteDlNode.addEventListener('click', (event) => {
	const clickedEl = event.target;
	if (clickedEl !== selectedAnnoteEl) {
	unselectCodeLines();
	const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
	if (activeEl) {
	activeEl.classList.remove('code-annotation-active');
	}
	selectCodeLines(clickedEl);
	clickedEl.classList.add('code-annotation-active');
	} else {
	// Unselect the line
	unselectCodeLines();
	clickedEl.classList.remove('code-annotation-active');
	}
	});
	}
	const findCites = (el) => {
	const parentEl = el.parentElement;
	if (parentEl) {
	const cites = parentEl.dataset.cites;
	if (cites) {
	return {
	el,
	cites: cites.split(' ')
	};
	} else {
	return findCites(el.parentElement)
	}
	} else {
	return undefined;
	}
	};
	var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
	for (var i=0; i<bibliorefs.length; i++) {
	const ref = bibliorefs[i];
	const citeInfo = findCites(ref);
	if (citeInfo) {
	tippyHover(citeInfo.el, function() {
	var popup = window.document.createElement('div');
	citeInfo.cites.forEach(function(cite) {
	var citeDiv = window.document.createElement('div');
	citeDiv.classList.add('hanging-indent');
	citeDiv.classList.add('csl-entry');
	var biblioDiv = window.document.getElementById('ref-' + cite);
	if (biblioDiv) {
	citeDiv.innerHTML = biblioDiv.innerHTML;
	}
	popup.appendChild(citeDiv);
	});
	return popup.innerHTML;
	});
	}
	}
	});
	</script>
	</div> <!-- /content -->




	</body></html>