1"use strict";(globalThis.webpackChunkmy_website=globalThis.webpackChunkmy_website||[]).push([[6194,54662,81049],{20924(e,a,t){t.r(a),t.d(a,{default:()=>o});var i=t(28774),n=(t(96540),t(74431)),s=t(74848);function o(){return(0,s.jsx)(s.Fragment,{children:(0,s.jsx)("div",{className:n.A.headerSecurity,children:(0,s.jsx)("div",{className:"container",children:(0,s.jsxs)("div",{className:n.A.box,children:[(0,s.jsx)("h1",{children:"Weaviate Knowledge Cards"}),(0,s.jsxs)("div",{className:n.A.headerBox,children:[(0,s.jsx)("p",{children:"Unlock the power of vector search. Our guides will help you conquer vector embeddings and build better AI applications."}),(0,s.jsx)("div",{className:n.A.buttons,children:(0,s.jsx)(i.A,{to:"/learn/knowledgecards",className:n.A.buttonGradient,children:"Back to Knowledge Cards"})})]})]})})})})}},35542(e,a,t){t.r(a),t.d(a,{default:()=>o});var i=t(96540),n=t(36228),s=(t(5260),t(74848));const o=({url:e})=>{const[a,t]=(0,i.useState)(!1),[o,r]=(0,i.useState)(!1),d=e=>{window.open(e,"_blank"),t(!1),r(!1)};return(0,s.jsx)(s.Fragment,{children:(0,s.jsxs)("div",{className:n.A.shareContainer,children:[!o&&!a&&(0,s.jsx)("button",{className:n.A.shareButton,onClick:()=>t(!0),children:"Share"}),o&&(0,s.jsx)("div",{className:n.A.shareButton,children:(0,s.jsx)("svg",{xmlns:"http://www.w3.org/2000/svg",width:"24",height:"24",viewBox:"0 0 24 24",children:(0,s.jsx)("path",{d:"M12 2c5.514 0 10 4.486 10 10s-4.486 10-10 10-10-4.486-10-10 4.486-10 10-10zm0-2c-6.627 0-12 5.373-12 12s5.373 12 12 12 12-5.373 12-12-5.373-12-12-12zm-1.959 17l-4.5-4.319 1.395-1.435 3.08 2.937 7.021-7.183 1.422 1.409-8.418 8.591z"})})}),a&&!o&&(0,s.jsxs)("div",{className:`${n.A.shareButton} ${n.A.shareIcons}`,children:[(0,s.jsx)("div",{onClick:()=>d(`https://twitter.com/intent/tweet?url=${encodeURIComponent(e)}`),children:(0,s.jsx)("svg",{xmlns:"http://www.w3.org/2000/svg",width:"24",height:"24",viewBox:"0 0 24 24",children:(0,s.jsx)("path",{d:"M 14.2856 10.172 L 23.2216 0 h -2.1168 l -7.7624 8.8304 L 7.1472 0 H 0 l 9.3704 13.3544 L 0 24.02 h 2.1168 l 8.192 -9.3272 l 6.544 9.3272 h 7.1472 M 2.8808 1.5632 H 6.1328 l 14.9704 20.9704 h -3.2528"})})}),(0,s.jsx)("div",{onClick:()=>d(`https://www.linkedin.com/sharing/share-offsite/?url=${encodeURIComponent(e)}`),children:(0,s.jsx)("svg",{xmlns:"http://www.w3.org/2000/svg",width:"24",height:"24",viewBox:"0 0 24 24",children:(0,s.jsx)("path",{d:"M0 0v24h24v-24h-24zm8 19h-3v-11h3v11zm-1.5-12.268c-.966 0-1.75-.79-1.75-1.764s.784-1.764 1.75-1.764 1.75.79 1.75 1.764-.783 1.764-1.75 1.764zm13.5 12.268h-3v-5.604c0-3.368-4-3.113-4 0v5.604h-3v-11h3v1.765c1.397-2.586 7-2.777 7 2.476v6.759z"})})}),(0,s.jsxs)("div",{onClick:async()=>{try{await navigator.clipboard.writeText(e),t(!1),r(!0)}catch(a){console.error("Failed to copy: ",a)}},children:[" ",(0,s.jsx)("svg",{width:"24",height:"24",xmlns:"http://www.w3.org/2000/svg","fill-rule":"evenodd","clip-rule":"evenodd",children:(0,s.jsx)("path",{d:"M14.851 11.923c-.179-.641-.521-1.246-1.025-1.749-1.562-1.562-4.095-1.563-5.657 0l-4.998 4.998c-1.562 1.563-1.563 4.095 0 5.657 1.562 1.563 4.096 1.561 5.656 0l3.842-3.841.333.009c.404 0 .802-.04 1.189-.117l-4.657 4.656c-.975.976-2.255 1.464-3.535 1.464-1.28 0-2.56-.488-3.535-1.464-1.952-1.951-1.952-5.12 0-7.071l4.998-4.998c.975-.976 2.256-1.464 3.536-1.464 1.279 0 2.56.488 3.535 1.464.493.493.861 1.063 1.105 1.672l-.787.784zm-5.703.147c.178.643.521 1.25 1.026 1.756 1.562 1.563 4.096 1.561 5.656 0l4.999-4.998c1.563-1.562 1.563-4.095 0-5.657-1.562-1.562-4.095-1.563-5.657 0l-3.841 3.841-.333-.009c-.404 0-.802.04-1.189.117l4.656-4.656c.975-.976 2.256-1.464 3.536-1.464 1.279 0 2.56.488 3.535 1.464 1.951 1.951 1.951 5.119 0 7.071l-4.999 4.998c-.975.976-2.255 1.464-3.535 1.464-1.28 0-2.56-.488-3.535-1.464-.494-.495-.863-1.067-1.107-1.678l.788-.785z"})})]})]})]})})}},64049(e,a,t){t.r(a),t.d(a,{default:()=>m});var i=t(96540),n=t(62891),s=t(5260),o=t(28774),r=t(39809),d=t(36228),l=t(20924),c=t(35542),g=t(38193),h=t(74848);const m=()=>{r.Q.length;const e=r.Q.find(e=>"Variable Dimensions"===e.title);if(!e)return(0,h.jsx)("p",{children:"Card not found"});const a="https://weaviate.io",t=(e.cardImage&&e.cardImage,e=>e.replace(/[^a-zA-Z0-9\s]/g,"").trim().replace(/\s+/g,"-").toLowerCase());let m="";g.A.canUseDOM&&(m=`${window.location.origin}/learn/knowledgecards/${t(e.title)}`);const u=e.type?e.type.toLowerCase():"",p=(t(e.title),r.Q.filter(a=>a.category===e.category)),v=p.length,[w,b]=(0,i.useState)(p.findIndex(a=>a.title===e.title));return(0,h.jsx)("div",{className:"custom-page noBG",children:(0,h.jsxs)(n.A,{children:[(0,h.jsxs)(s.A,{children:[(0,h.jsxs)("title",{children:[e.title," - Weaviate Knowledge Cards"]}),(0,h.jsx)("meta",{property:"og:title",content:"Variable Dimensions - Weaviate Knowledge Cards"}),(0,h.jsx)("meta",{name:"robots",content:"index, follow"}),(0,h.jsx)("meta",{property:"og:type",content:"article"}),(0,h.jsx)("meta",{property:"og:image",content:"https://weaviate.io/img/cards/variable-dimensions.jpg"}),(0,h.jsx)("meta",{property:"og:image:alt",content:"Knowledge c
1ard image"}),(0,h.jsx)("meta",{property:"og:url",content:"https://weaviate.io/learn/knowledgecards/variable-dimensions"}),(0,h.jsx)("meta",{property:"og:description",content:"Flexible embedding sizes, like Matryoshka embeddings. Encode information hierarchically, allowing adaptation..."}),(0,h.jsx)("meta",{property:"og:site_name",content:"Weaviate Knowledge Cards"}),(0,h.jsx)("meta",{property:"og:locale",content:"en_US"}),(0,h.jsx)("meta",{property:"article:published_time",content:"2024-06-07T12:35+00:00"}),(0,h.jsx)("meta",{property:"article:author",content:"weaviate.io"}),(0,h.jsx)("meta",{name:"twitter:card",content:"summary_large_image"}),(0,h.jsx)("meta",{name:"twitter:site",content:"@weaviate_io"}),(0,h.jsx)("meta",{name:"twitter:title",content:"Variable Dimensions - Weaviate Knowledge Cards"}),(0,h.jsx)("meta",{name:"twitter:description",content:"Flexible embedding sizes, like Matryoshka embeddings. Encode information hierarchically, allowing adaptation..."}),(0,h.jsx)("meta",{name:"twitter:image",content:"https://weaviate.io/img/cards/variable-dimensions.jpg"}),(0,h.jsx)("meta",{name:"twitter:image:alt",content:"Knowledge card image"}),(0,h.jsx)("script",{type:"application/ld+json",children:JSON.stringify({"@context":"http://schema.org","@type":"Article",headline:"Variable Dimensions- Weaviate Knowledge Cards",description:"Flexible embedding sizes, like Matryoshka embeddings. Encode information hierarchically, allowing adaptation...",image:"https://weaviate.io/img/cards/variable-dimensions.jpg",url:"https://weaviate.io/learn/knowledgecards/variable-dimensions"})})]}),(0,h.jsx)(l.default,{}),(0,h.jsxs)("main",{className:d.A.cardPage,children:[(0,h.jsxs)("div",{className:d.A.breadCrumbs,children:[(0,h.jsx)(o.A,{to:"/learn/knowledgecards",children:"Knowledge Cards"}),(0,h.jsx)("span",{children:" / "}),(0,h.jsx)("span",{children:e.category}),(0,h.jsx)("span",{children:" / "}),(0,h.jsx)("span",{children:e.title})]}),(0,h.jsx)("div",{className:d.A.modals,children:(0,h.jsxs)("div",{className:`${d.A.modalContents} ${d.A[u]}`,children:[(0,h.jsx)("div",{className:d.A.cardHeader,children:(0,h.jsx)("span",{className:d.A.cardType,children:e.category})}),(0,h.jsxs)("div",{className:d.A.cardContents,children:[e.photo&&(0,h.jsx)("img",{src:`/img/site/${e.photo}`,alt:e.title,className:d.A.cardImage}),(0,h.jsx)("h3",{className:d.A.cardTitle,children:e.title}),(0,h.jsx)("p",{className:d.A.modalText,children:e.longText}),(0,h.jsx)(c.default,{url:m}),(0,h.jsx)("div",{className:d.A.bottomCard,children:(e.bloglink||e.bloglink2||e.doclink||e.doclink2||e.videolink)&&(0,h.jsxs)(h.Fragment,{children:[(0,h.jsx)("p",{className:d.A.relatedText,children:"Related Content:"}),(0,h.jsxs)("div",{className:d.A.relatedBox,children:[e.bloglink&&(0,h.jsx)(o.A,{to:e.bloglink,children:(0,h.jsxs)("div",{className:d.A.relatedBlog,children:[(0,h.jsx)("div",{className:d.A.relatedImage}),(0,h.jsxs)("div",{className:d.A.relatedBottom,children:[(0,h.jsx)("span",{className:d.A.relatedTitle,children:"Blog:"}),(0,h.jsx)("span",{className:d.A.relatedSubtitle,children:e.blogTitle})]})]})}),e.bloglink2&&(0,h.jsx)(o.A,{to:e.bloglink2,children:(0,h.jsxs)("div",{className:d.A.relatedBlog,children:[(0,h.jsx)("div",{className:d.A.relatedImage}),(0,h.jsxs)("div",{className:d.A.relatedBottom,children:[(0,h.jsx)("span",{className:d.A.relatedTitle,children:"Blog:"}),(0,h.jsx)("span",{className:d.A.relatedSubtitle,children:e.blogTitle2})]})]})}),e.doclink&&(0,h.jsx)(o.A,{to:e.doclink,children:(0,h.jsxs)("div",{className:d.A.relatedBlog,children:[(0,h.jsx)("div",{className:d.A.relatedImage}),(0,h.jsxs)("div",{className:d.A.relatedBottom,children:[(0,h.jsx)("span",{className:d.A.relatedTitle,children:"Doc:"}),(0,h.jsx)("span",{className:d.A.relatedSubtitle,children:e.docTitle})]})]})}),e.doclink2&&(0,h.jsx)(o.A,{to:e.doclink2,children:(0,h.jsxs)("div",{className:d.A.relatedBlog,children:[(0,h.jsx)("div",{className:d.A.relatedImage}),(0,h.jsxs)("div",{className:d.A.relatedBottom,children:[(0,h.jsx)("span",{className:d.A.relatedTitle,children:"Doc:"}),(0,h.jsx)("span",{className:d.A.relatedSubtitle,children:e.docTitle2})]})]})}),e.videolink&&(0,h.jsx)(o.A,{to:e.videolink,children:(0,h.jsxs)("div",{className:d.A.relatedBlog,children:[(0,h.jsx)("div",{className:d.A.relatedImage}),(0,h.jsxs)("div",{className:d.A.relatedBottom,children:[(0,h.jsx)("span",{className:d.A.relatedTitle,children:"Video:"}),(0,h.jsx)("span",{className:d.A.relatedSubtitle,children:e.videoTitle})]})]})})]})]})}),(0,h.jsx)("span",{className:d.A.nextText,children:`${w+1} of ${v}`}),(0,h.jsxs)("div",{className:d.A.nextContainer,children:[(0,h.jsx)(o.A,{to:e.previous,className:d.A.nextButton,children:"Previous"}),(0,h.jsx)(o.A,{to:e.next,className:d.A.nextButton,children:"Next"})]})]})]})})]})]})})}},74431(e,a,t){t.d(a,{A:()=>i});const i={headerSecurity:"headerSecurity_xxlA",box:"box_JiK5",headerBox:"headerBox_OBQR",searchContainer:"searchContainer_eQeh",buttons:"buttons_svIQ",buttonGradient:"buttonGradient_aPfa"}},36228(e,a,t){t.d(a,{A:()=>i});const i={cardPage:"cardPage_zDfq",breadCrumbs:"breadCrumbs_XNyt",modals:"modals_EBY0",modalContents:"modalContents_KyK8",cardHeader:"cardHeader_S9Ff",cardType:"cardType_gg2z",cardContents:"cardContents_MsYV",cardImage:"cardImage_hB4R",cardTitle:"cardTitle_u4MT",modalText:"modalText_em9R",shareContainer:"shareContainer_acBY",shareButton:"shareButton_Mzj6",shareIcons:"shareIcons_vif1",bottomCard:"bottomCard_vaF7",relatedText:"relatedText_tox0",relatedBox:"relatedBox_ANl_",relatedBlog:"relatedBlog_hGRD",relatedImage:"relatedImage_CzaS",relatedBottom:"relatedBottom_Cbi0",relatedTitle:"relatedTitle_MYSo",relatedSubtitle:"relatedSubtitle_UgrT",nextText:"nextText_rz9k",nextContainer:"nextContainer_faYg",nextButton:"nextButton_wBIv",introduction:"introduction_EDXR",search:"search_gbnJ",hnsw:"hnsw_htcH",rag:"rag_6upP",databases:"databases_eIVK",llms:"llms_JYm5",et:"et_jXqJ",ct:"ct_NL6G"}},39809(e){e.exports=JSON.parse('{"Q":[{"id":"vector-databases-1","type":"Introduction","category":"Intro to Vector Databases","categoryText":"One sentence about the category","title":"Unstructured Data Objects","photo":"unstructured-data-objects-card-icon.svg","text":"Unstructured data objects are data objects without a predefined structure, making them difficult to manage conventionally. Examples include text documents, images, audio and video...","longText":"Unstructured data objects are data objects without a predefined structure, making them difficult to manage conventionally. Examples include text documents, images, audio, video, social media posts, and emails.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/understand-your-unstructured-data#intro","blogTitle":"Understand Your Unstructured Data","bloglink2":"https://weaviate.io/blog/the-ai-first-database-ecosystem#were-awash-in-unstructured-data","blogTitle2":"The AI-First Database Ecosystem","tags":["Vector Databases","Weaviate","Image"],"cardImage":"unstructured-data-objects.jpg","next":"/learn/knowledgecards/vector","previous":"/learn/knowledgecards/hnsw"},{"id":"vector-databases-2","type":"Introduction","category":"Intro to Vector Databases","title":"Vector","photo":"vector-card-icon.svg","text":"Vectors, or vector embeddings in databases, are quantities with magnitude and direction...","longText":"Vectors, or vector embeddings in databases, are quantities with magnitude and direction represented by an array of numerical values. They represent data objects in multi-dimensional space.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/search/similarity#search-with-a-vector","docT
1itle":"Search with a Vector","tags":["Vector Databases","Weaviate","Image"],"cardImage":"vector.jpg","next":"/learn/knowledgecards/vector-embedding","previous":"/learn/knowledgecards/unstructured-data-objects"},{"id":"vector-databases-3","type":"Introduction","category":"Intro to Vector Databases","title":"Vector Embedding","photo":"vector-embedding-card-icon.svg","text":"Numerical representations of objects, such as words or images, in a vector space...","longText":"Numerical representations of objects, such as words or images, in a vector space. These representations capture semantic relationships and are used in machine-learning tasks.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-embeddings-explained","blogTitle":"Vector Embeddings Explained","bloglink2":"https://weaviate.io/blog/distance-metrics-in-vector-search#vectors-in-multidimensional-space","blogTitle2":"Distance Metrics in Vector Search","tags":["Vector Databases","Weaviate","Image"],"cardImage":"vector-embedding.jpg","next":"/learn/knowledgecards/embeddings-model","previous":"/learn/knowledgecards/vector"},{"id":"vector-databases-4","type":"Introduction","category":"Intro to Vector Databases","title":"Embeddings Model","photo":"embed-model-card-icon.svg","text":"An algorithm or model that transforms objects (like words) into vector embeddings...","longText":"An algorithm or model that transforms objects (like words) into vector embeddings. This enables capturing contextual and semantic information.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-embeddings-explained","blogTitle":"Vector Embeddings Explained","doclink":"https://docs.weaviate.io/weaviate/model-providers","docTitle":"Retriever Vectorizer Modules","videolink":"https://www.youtube.com/watch?v=lpdN3aw-yTg","videoTitle":"Vector embedding models for AI","tags":["Vector Databases","Weaviate","Image"],"cardImage":"embeddings-model.jpg","next":"/learn/knowledgecards/vectorbased-index","previous":"/learn/knowledgecards/vector-embedding"},{"id":"vector-databases-5","type":"Introduction","category":"Intro to Vector Databases","title":"Vector-Based Index","photo":"vector-index-card-icon.svg","text":"A vector index is a type of data structure for efficiently storing high-dimensional vector data...","longText":"A vector index is a type of data structure for efficiently storing high-dimensional vector data generated by machine learning models. Its main purpose is facilitating quick and accurate similarity searches, using algorithms like trees or hashing to reduce complexity in high-dimensional spaces.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/what-is-a-vector-database#vector-indexing-for-approximate-nearest-neighbor-approach","blogTitle":"What is a Vector Database?","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-index","docTitle":"Vector Index","tags":["Vector Databases","Weaviate","Image"],"cardImage":"vector-based-index.jpg","next":"/learn/knowledgecards/vector-database","previous":"/learn/knowledgecards/embeddings-model"},{"id":"vector-databases-6","type":"Introduction","category":"Intro to Vector Databases","title":"Vector Database","photo":"vector-database-card-icon.svg","text":"A vector database indexes, stores, and provides access to structured or unstructured data...","longText":"A vector database indexes, stores, and provides access to structured or unstructured data (e.g., text or images) alongside its\xa0vector embeddings, which are the data\'s numerical representation. It allows users to find and retrieve similar objects quickly at scale in production.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-library-vs-vector-database#vector-databases","blogTitle":"Vector Library vs Vector Database","doclink":"https://weaviate.io/blog/vector-library-vs-vector-database#vector-databases","docTitle":"Vector Library vs Vector Database","tags":["Vector Databases","Weaviate","Image"],"cardImage":"vector-database.jpg","next":"/learn/knowledgecards/hnsw","previous":"/learn/knowledgecards/vectorbased-index"},{"id":"vector-databases-7","type":"Introduction","category":"Intro to Vector Databases","title":"HNSW","photo":"HNSW-card-icon.svg","text":"HNSW (Hierarchical Navigable Small World) is a graph-based vector indexing method for...","longText":"HNSW (Hierarchical Navigable Small World) is a graph-based vector indexing method for efficient approximate nearest neighbor searches. It organizes data in a multi-layer structure, enabling fast and accurate queries by navigating through layers from coarse to fine granularity.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/ann-algorithms-hnsw-pq","blogTitle":"HNSW+PQ - Exploring ANN algorithms Part 2.1","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-index#what-is-hierarchical-navigable-small-world-hnsw","docTitle":"Vector Indexing","tags":["Vector Databases","Weaviate","HNSW"],"cardImage":"hnsw.jpg","next":"/learn/knowledgecards/unstructured-data-objects","previous":"/learn/knowledgecards/vector-database"},{"id":"search-1","type":"Search","category":"Search","title":"Sparse Vectors","photo":"hybrid-search-icon.svg","text":"Sparse embeddings are generated from algorithms like BM25 and SPLADE...","longText":"Sparse vectors are often high-dimensional with many zeros. Sparse embeddings are generated from algorithms like BM25 and SPLADE. These algorithms are used in keyword-based search to represent text documents.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained#sparse-and-dense-vectors","blogTitle":"Hybrid Search Explained","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"sparse-vectors.jpg","next":"/learn/knowledgecards/dense-vectors","previous":"/learn/knowledgecards/semanticvector-search"},{"id":"search-2","type":"Search","category":"Search","title":"Dense Vectors","photo":"dense-vectors-card-icon.svg","text":"In contrast to sparse vectors, dense vectors contain mostly non-zero values and are generated from machine learning models like GloVe and Transformers...","longText":"In contrast to sparse vectors, dense vectors contain mostly non-zero values and are generated from machine learning models like GloVe and Transformers. These vectors capture the semantic meaning of text and are used in vector search techniques.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained#sparse-and-dense-vectors","blogTitle":"Hybrid Search Explained","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"dense-vectors.jpg","next":"/learn/knowledgecards/bm25bm25f","previous":"/learn/knowledgecards/sparse-vectors"},{"id":"search-3","type":"Search","category":"Search","title":"BM25/BM25F","photo":"bm25-card-icon.svg","text":"BM25 is a ranking function used by search engines to estimate the relevance of documents to a given search query. It is part of the family of probabilistic information retrieval models...","longText":"BM25 is a ranking function used by search engines to estimate the relevance of documents to a given search query. It is part of the family of probabilistic information retrieval models. BM25F is a variant of BM25 that can handle fields with different weights, such as title and body in documents.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained#bm25","blogTitle":"Hybrid Search Explained","doclink":"https://docs.weaviate.io/weaviate/api/gra
1phql/search-operators#bm25","docTitle":"Search Operators","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"BM25-BM25F.jpg","next":"/learn/knowledgecards/alpha-parameter","previous":"/learn/knowledgecards/dense-vectors"},{"id":"search-4","type":"Search","category":"Search","title":"Alpha Parameter","photo":"alpha-parameter-card-icon.svg","text":"This is an optional parameter in hybrid search queries that determines the weighting of each search algorithm during the re-ranking process...","longText":"This is an optional parameter in hybrid search queries that determines the weighting of each search algorithm during the re-ranking process. An alpha value of 0 gives full weight to the sparse vector (BM25), 1 gives full weight to the dense vector, and 0.5 gives equal weight to both.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-for-web-developers#changing-alpha","blogTitle":"Hybrid Search for Web Developers","doclink":"https://docs.weaviate.io/weaviate/search/hybrid#balance-keyword-and-vector-search","docTitle":"Balance Keyword and Vector Search","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"alpha-parameter.jpg","next":"/learn/knowledgecards/fusion-algorithm","previous":"/learn/knowledgecards/bm25bm25f"},{"id":"search-5","type":"Search","category":"Search","title":"Fusion Algorithm","photo":"fusion-algorithm-card-icon.svg","text":"In hybrid search, the fusion algorithm combines results from keyword and vector searches into a final list of ranked search results...","longText":"In hybrid search, the fusion algorithm combines results from keyword and vector searches into a final list of ranked search results. A popular fusion algorithm is called reciprocal ranked fusion, which uses the sum of the inverse of the position of the results. There are multiple types of fusion algorithms. Weaviate supports reciprocal rank fusion and relative score fusion.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-fusion-algorithms","blogTitle":"Hybrid Search Fusion Algorithms","doclink":"https://docs.weaviate.io/weaviate/search/hybrid#change-the-fusion-method","docTitle":"Change the Fusion Method","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"fusion-algorithm.jpg","next":"/learn/knowledgecards/hybrid-search","previous":"/learn/knowledgecards/alpha-parameter"},{"id":"search-6","type":"Search","category":"Search","title":"Hybrid Search","photo":"hybrid-search-card-icon.svg","text":"Hybrid search in Weaviate combines keyword (BM25) and vector search to leverage both exact term matching and semantic context...","longText":"Hybrid search in Weaviate combines keyword (BM25) and vector search to leverage both exact term matching and semantic context. It uses rankedFusion or relativeScoreFusion algorithms to merge results, providing a balance between precision and semantic understanding.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained#when-to-use-hybrid-search","blogTitle":"When to Use Hybrid Search?","bloglink2":"https://weaviate.io/blog/hybrid-search-fusion-algorithms#base-search-results","blogTitle2":"Base Search Results","doclink":"https://docs.weaviate.io/weaviate/search/hybrid","docTitle":"Hybrid Search","tags":["Hybrid Search","Weaviate","Image"],"cardImage":"hybrid-search.jpg","next":"/learn/knowledgecards/keyword-search","previous":"/learn/knowledgecards/fusion-algorithm"},{"id":"search-7","type":"Search","category":"Search","title":"Keyword Search","photo":"keyword-search-card-icon.svg","text":"Keyword search finds objects with an exact match between the terms in the query and in the documents...","longText":"Keyword search finds objects with an exact match between the terms in the query and in the documents. Weaviate uses the BM25/F function to calculate the relevancy of a document to a query, which uses both the term frequency and inverse document frequency.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained","blogTitle":"Hybrid Search Explained","bloglink2":"https://weaviate.io/blog/hybrid-search-for-web-developers","blogTitle2":"A Web Developers Guide to Hybrid Search","doclink":"https://docs.weaviate.io/weaviate/search/bm25","docTitle":"Keyword Search","tags":["Keyword Search","Weaviate","Image"],"cardImage":"keyword-search.jpg","next":"/learn/knowledgecards/semanticvector-search","previous":"/learn/knowledgecards/hybrid-search"},{"id":"search-8","type":"Search","category":"Search","title":"Semantic/Vector Search","photo":"semantic-vector-search-card-icon.svg","text":"Semantic search, also known as vector search, uses machine learning to grasp text context, not just keywords...","longText":"Semantic search, also known as vector search, uses machine learning to grasp text context, not just keywords. It converts text to numerical vectors, finding matches based on conceptual similarity for accurate, relevant search results.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/healthsearch-demo","blogTitle":"Unlocking Health with Semantic Search","doclink":"https://docs.weaviate.io/weaviate/search/similarity","docTitle":"Vector similarity search","tags":["Keyword Search","Weaviate","Semantic Search"],"cardImage":"semantic-vector-search.jpg","next":"/learn/knowledgecards/sparse-vectors","previous":"/learn/knowledgecards/keyword-search"},{"id":"hnsw-1","type":"HNSW","category":"Hierarchical Navigable Small World","title":"Hierarchical Graph Structure","photo":"hgs-card-icon.svg","text":"HNSW utilizes a multi-layered graph structure where each layer is a graph of the data points. The top layers contain fewer points with long-range connections...","longText":"HNSW utilizes a multi-layered graph structure where each layer is a graph of the data points. The top layers contain fewer points with long-range connections, and as you go down the layers, the number of points increases, and the connections become more local. This hierarchical structure allows for efficient navigation during search queries.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/ann-algorithms-vamana-vs-hnsw#hnsw-indexing--in-short","blogTitle":"ANN Algorithms: Vamana vs HNSW","bloglink2":"https://weaviate.io/blog/ann-algorithms-hnsw-pq#what-information-to-move-to-disk","blogTitle2":"ANN Algorithms: HNSW & PQ","doclink":"https://docs.weaviate.io/weaviate/introduction#how-does-weaviate-work","docTitle":"How Does Weaviate Work?","tags":["HNSW","Weaviate","Image"],"cardImage":"hierarchical-graph-structure.jpg","next":"/learn/knowledgecards/ann-approximate-nearest-neighbor","previous":"/learn/knowledgecards/graphbased-index"},{"id":"hnsw-2","type":"HNSW","category":"Hierarchical Navigable Small World","title":"ANN - Approximate Nearest Neighbor","photo":"ann-card-icon.svg","text":"ANN algorithms calculate the approximate nearest neighbors to a query, as opposed to kNN algorithms, which calculate the true nearest neighbors...","longText":"ANN algorithms calculate the approximate nearest neighbors to a query, as opposed to kNN algorithms, which calculate the true nearest neighbors. ANN algorithms enable the quick and efficie
1nt identification of the closest data points at scale. This makes ANN suitable for big data where speed matters more than perfect accuracy.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-search-explained","blogTitle":"Why is Vector Search so Fast?","bloglink2":"https://weaviate.io/blog/crud-support-in-weaviate","blogTitle2":"CRUD Support in Weaviate","doclink":"https://docs.weaviate.io/weaviate/concepts/indexing#ann-index","docTitle":"ANN Indexing","tags":["HNSW","Weaviate","Image"],"cardImage":"ann-approximate-nearest-neighbor.jpg","next":"/learn/knowledgecards/layered-navigation","previous":"/learn/knowledgecards/hierarchical-graph-structure"},{"id":"hnsw-3","type":"HNSW","category":"Hierarchical Navigable Small World","title":"Layered Navigation","photo":"layered-navigation-card-icon.svg","text":"During a search, HNSW starts at the topmost layer and makes \'big jumps\' across the graph to quickly move closer to the target region...","longText":"During a search, HNSW starts at the topmost layer and makes \'big jumps\' across the graph to quickly move closer to the target region. As it moves down the layers, the search becomes more refined, with smaller, more local jumps until the nearest neighbors are identified in the bottom layer.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-index#hierarchical-navigable-small-world-hnsw-index","docTitle":"HNSW Index","tags":["HNSW","Weaviate","Image"],"cardImage":"layered-navigation.jpg","next":"/learn/knowledgecards/graphbased-index","previous":"/learn/knowledgecards/ann-approximate-nearest-neighbor"},{"id":"hnsw-4","type":"HNSW","category":"Hierarchical Navigable Small World","title":"Graph-Based Index","photo":"graph-based-card-icon.svg","text":"HNSW is a graph-based indexing technique, meaning that it organizes the data points in a graph structure rather than traditional tree-based or hash-based structures...","longText":"HNSW is a graph-based indexing technique, meaning that it organizes the data points in a graph structure rather than traditional tree-based or hash-based structures. This graph-based approach is key to its performance in high-dimensional vector space searches.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-search-explained","blogTitle":"Why is Vector Search so Fast?","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-index#hierarchical-navigable-small-world-hnsw-index","docTitle":"HNSW Index","videolink":"","videoTitle":"","tags":["HNSW","Weaviate","Image"],"cardImage":"graph-based-index.jpg","next":"/learn/knowledgecards/hierarchical-graph-structure","previous":"/learn/knowledgecards/layered-navigation"},{"id":"rag-1","type":"RAG","category":"Multimodal RAG","title":"Cross Modal Reasoning","photo":"cross-card-icon.svg","text":"Cross-modal reasoning refers to the ability to make connections by integrating information from different modalities or sources including text, images, audio and video...","longText":"Cross-modal reasoning refers to the ability to make connections by integrating information from different modalities or sources including text, images, audio and video. It involves leveraging knowledge from one modality to understand or reason about data from another modality.","link":"https://docs.weaviate.io/weaviate/quickstart","tags":["RAG","Weaviate","Image"],"cardImage":"cross-modal-reasoning.jpg","next":"/learn/knowledgecards/multimodal-embeddings-models","previous":"/learn/knowledgecards/multimodal-rag"},{"id":"rag-2","type":"RAG","category":"Multimodal RAG","title":"Multimodal Embeddings Models","photo":"multimodal-card-icon.svg","text":"Multimodal Embeddings Models produce a joint embedding space for multimodal data that understands text, images, audio and more...","longText":"Multimodal Embeddings Models produce a joint embedding space for multimodal data that understands text, images, audio and more. Objects that are similar are closer together and dissimilar objects are farther apart, this means that the model preserves semantic similarity within and across modalities.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/multimodal-models","blogTitle":"Multimodal Embedding Models","tags":["RAG","Weaviate","Image"],"cardImage":"multimodal-embeddings-models.jpg","next":"/learn/knowledgecards/multimodal-contrastive-finetuning","previous":"/learn/knowledgecards/cross-modal-reasoning"},{"id":"rag-3","type":"RAG","category":"Multimodal RAG","title":"Multimodal Contrastive Finetuning","photo":"contrastive-card-icon.svg","text":"Multimodal Embeddings Models produce a joint embedding space for multimodal data that understands text, images, audio and more...","longText":"Multimodal Embeddings Models produce a joint embedding space for multimodal data that understands text, images, audio and more. Objects that are similar are closer together and dissim
1ilar objects are farther apart, this means that the model preserves semantic similarity within and across modalities.","link":"https://docs.weaviate.io/weaviate/quickstart","tags":["RAG","Weaviate","Image"],"cardImage":"multimodal-contrastive-finetuning.jpg","next":"/learn/knowledgecards/anytoany-search","previous":"/learn/knowledgecards/multimodal-embeddings-models"},{"id":"rag-4","type":"RAG","category":"Multimodal RAG","title":"Any-to-any Search","photo":"any-search-card-icon.svg","text":"In any-to-any search we can pass in as a query any modality the model understands and use it to perform vector similarity search...","longText":"In any-to-any search we can pass in as a query any modality the model understands and use it to perform vector similarity search in multimodal embedding space, getting back objects of any other modality that are similar in concept.","link":"https://docs.weaviate.io/weaviate/quickstart","tags":["RAG","Weaviate","Image"],"cardImage":"any-to-any-search.jpg","next":"/learn/knowledgecards/multimodal-rag","previous":"/learn/knowledgecards/multimodal-contrastive-finetuning"},{"id":"rag-5","type":"RAG","category":"Multimodal RAG","title":"Multimodal RAG","photo":"rag-card-icon.svg","text":"Multimodal RAG involves retrieving from a multimodal knowledge base and then generation using a large multimodal model by...","longText":"Multimodal RAG involves retrieving from a multimodal knowledge base and then generation using a large multimodal model by generating text or images grounded in the retrieved context, which can include images, text, audio, and other modalities.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/multimodal-rag#multimodal-retrieval-augmented-generationmmrag","blogTitle":"Multimodal Retrieval Augmented Generation","doclink":"https://docs.weaviate.io/weaviate/api/graphql/search-operators#multimodal-search","docTitle":"Multimodal Search","tags":["RAG","Weaviate","Image"],"cardImage":"multimodal-rag.jpg","next":"/learn/knowledgecards/cross-modal-reasoning","previous":"/learn/knowledgecards/anytoany-search"},{"id":"databases-1","type":"Databases","category":"Databases","title":"Graph Database","photo":"graph-database-icon.svg","text":"A graph database stores data in nodes and edges, representing entities and their relationships. It\'s optimized for...","longText":"A graph database stores data in nodes and edges, representing entities and their relationships. It\'s optimized for querying connections and patterns within the data, making it powerful for network, social, and recommendation systems.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/more-resources/performance#cost-of-resolving-referencing","docTitle":"Cost of resolving referencing","tags":["Databases","Weaviate","Image"],"cardImage":"graph-database.jpg","next":"/learn/knowledgecards/inverted-indexes","previous":"/learn/knowledgecards/pq"},{"id":"databases-2","type":"Databases","category":"Databases","title":"Inverted Indexes","photo":"inverted-indexes-icon.svg","text":"An inverted index is a database indexing structure that maps keywords to their locations in documents...","longText":"An inverted index is a database indexing structure that maps keywords to their locations in documents, enabling fast full-text searches by listing all documents containing a given word, optimizing retrieval times. In Weaviate, it\'s used for keyword searches and fast filtering.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/more-resources/performance#inverted-indexes","docTitle":"Inverted index","doclink2":"https://docs.weaviate.io/weaviate/concepts/indexing#inverted-indexes","docTitle2":"Indexing","tags":["Databases","Weaviate","Image"],"cardImage":"inverted-indexes.jpg","next":"/learn/knowledgecards/sharding","previous":"/learn/knowledgecards/graph-database"},{"id":"databases-3","type":"Databases","category":"Databases","title":"Sharding","photo":"sharding-icon.svg","text":"Sharding is splitting a database into smaller, faster, more easily managed parts called shards...","longText":"Sharding is splitting a database into smaller, faster, more easily managed parts called shards. Each shard is a self-contained unit with a subset of data, allowing for distribution across servers or data centers for improved performance and scalability.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/starter-guides/managing-collections#sharding","docT
1itle":"Sharding","doclink2":"https://docs.weaviate.io/weaviate/concepts/replication-architecture#replication-vs-sharding","docTitle2":"Replication vs Sharding","tags":["Databases","Weaviate","Image"],"cardImage":"sharding.jpg","next":"/learn/knowledgecards/multi-tenancy","previous":"/learn/knowledgecards/inverted-indexes"},{"id":"databases-4","type":"Databases","category":"Databases","title":"Multi Tenancy","photo":"multi-tenancy-icon.svg","text":"With multi-tenancy, a single database instance serves multiple clients or \'tenants\'...","longText":"With multi-tenancy, a single database instance serves multiple clients or \'tenants\' simultaneously, while ensuring that each tenant\'s data remains isolated and secure. It also enables efficient resource use and cost savings over large datasets at scale.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/multi-tenancy-vector-search","blogTitle":"Multi-tenancy and Vector Search","doclink":"https://docs.weaviate.io/weaviate/manage-data/read-all-objects#read-all-objects---multi-tenant-collections","docTitle":"Read All Objects - Multi-tenant collections","tags":["Databases","Weaviate","Image"],"cardImage":"multi-tenancy.jpg","next":"/learn/knowledgecards/relational-database","previous":"/learn/knowledgecards/sharding"},{"id":"databases-5","type":"Databases","category":"Databases","title":"Relational Database","photo":"relational-database-icon.svg","text":"A relational database organizes data into tables linked by relationships...","longText":"A relational database organizes data into tables linked by relationships, enabling complex queries like finding students with both cats and dogs. It uses SQL for structured, precise data management, optimizing operations like filtering and joining for efficient data retrieval.","link":"https://docs.weaviate.io/weaviate/quickstart","tags":["Databases","Weaviate","Image"],"cardImage":"relational-database.jpg","next":"/learn/knowledgecards/pq","previous":"/learn/knowledgecards/multi-tenancy"},{"id":"databases-6","type":"Databases","category":"Databases","title":"PQ","photo":"pq-card-icon.svg","text":"Product quantization (PQ) is a compression technique for vectors, reducing memory usage by...","longText":"Product quantization (PQ) is a compression technique for vectors, reducing memory usage by up to 90%. It\'s a lossy algorithm, meaning it trades some accuracy for significant memory savings. However, Weaviate employs a rescoring technique to reduce the loss of recall.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/pq-rescoring","blogTitle":"PQ Rescoring","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-quantization#pq-compression-process","docTitle":"PQ compression process","tags":["Databases","Weaviate","Image"],"cardImage":"pq.jpg","next":"/learn/knowledgecards/graph-database","previous":"/learn/knowledgecards/relational-database"},{"id":"llms-1","type":"LLMS","category":"Large Language Models","title":"Large Language Model (LLM)","photo":"large-language-model-icon.svg","text":"A Large Language Model (LLM) is a machine learning model that is trained on vast text data, learning language...","longText":"A Large Language Model (LLM) is a machine learning model that is trained on vast text data, learning language patterns to understand, predict and generate human-like text. It statistically models language to uniquely generate responses to queries with learned context.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/llms-and-search","blogTitle":"LLMs and Search","bloglink2":"https://weaviate.io/blog/rag-evaluation#llm-evaluations","blogTitle2":"RAG Evaluation","doclink":"https://docs.weaviate.io/weaviate/starter-guides/generative#why-generative-search","docTitle":"Why generative search?","tags":["LLMs","Weaviate","Image"],"cardImage":"large-language-model.jpg","next":"/learn/knowledgecards/finetuning","previous":"/learn/knowledgecards/transformer-model"},{"id":"llms-2","type":"LLMS","category":"Large Language Models","title":"Finetuning","photo":"finetuning-icon.svg","text":"Fine-tuning is a process where a pre-trained machine learning model is further trained on a specific dataset...","longText":"Fine-tuning is a process where a pre-trained machine learning model is further trained on a specific dataset to specialize its knowledge or skills for a particular task or domain. This improves the model\'s performance on the related task while being more cost-efficient than training it from scratch.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/multimodal-models#2-model-architecture","blogTitle":"Multimodal Models","tags":["LLMs","Weaviate","Image"],"cardImage":"finetuning.jpg","next":"/learn/knowledgecards/multimodal","previous":"/learn/knowledgecards/large-language-model-llm"},{"id":"llms-3","type":"LLMS","category":"Large Language Models","title":"Multi-modal","photo":"multi-modal-icon.svg","text":"Multi-modal learning involves integrating various data types like images, text, audio, and sensory inputs...","longText":"Multimodal learning involves integrating various data types like images, text, audio, and sensory inputs to create models that understand and process information like humans, recognizing patterns and context across different modes for different types of use-cases, like e-commerce.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/multimodal-models","blogTitle":"Multimodal Embedding Models","bloglink2":"https://weaviate.io/blog/multimodal-rag","blogTitle2":"Multimodal RAG","doclink":"https://docs.weaviate.io/weaviate/api/gra
1phql/search-operators#multimodal-search","docTitle":"Multimodal Search","tags":["LLMs","Weaviate","Image"],"cardImage":"multi-modal.jpg","next":"/learn/knowledgecards/embedding-model","previous":"/learn/knowledgecards/finetuning"},{"id":"llms-4","type":"LLMS","category":"Large Language Models","title":"Embedding model","photo":"embedding-icon.svg","text":"An embedding model, or vectorizer, transforms text or images into vector embeddings, capturing semantic...","longText":"An embedding model, or vectorizer, transforms text or images into vector embeddings, capturing semantic meaning through statistical analysis, converting conceptual significance into numerical representations.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-embeddings-explained","blogTitle":"Vector Embeddings Explained","doclink":"https://docs.weaviate.io/weaviate/model-providers","docTitle":"Retriever Vectorizer Modules","videolink":"https://www.youtube.com/watch?v=lpdN3aw-yTg","videoTitle":"Vector embedding models for AI","tags":["LLMs","Weaviate","Image"],"cardImage":"embedding-model.jpg","next":"/learn/knowledgecards/chunking","previous":"/learn/knowledgecards/multimodal"},{"id":"llms-5","type":"LLMS","category":"Large Language Models","title":"Chunking","photo":"chunking-icon.svg","text":"Chunking is the process of dividing large texts into smaller, manageable pieces, called \'chunks\'...","longText":"Chunking is the process of dividing large texts into smaller, manageable pieces, called \'chunks\', for storage or processing in vector databases and language models. It improves search performance and the quality of language model outputs by defining the granularity of information retrieval.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/llms-and-search#index-construction","blogTitle":"LLMs and Search","bloglink2":"https://docs.weaviate.io/weaviate/starter-guides/generative#chunking","blogTitle2":"Chunking","tags":["LLMs","Weaviate","Image"],"cardImage":"chunking.jpg","next":"/learn/knowledgecards/generative-ai","previous":"/learn/knowledgecards/embedding-model"},{"id":"llms-6","type":"LLMS","category":"Large Language Models","title":"Generative AI","photo":"generative-ai-card-icon.svg","text":"Generative AI is a type of artificial intelligence powered by generative machine learning models that produces new...","longText":"Generative AI is a type of artificial intelligence powered by generative machine learning models that produces new and original outputs like images, text, audio, or video based off user provided prompts or input.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hurricane-generative-feedback-loops","blogTitle":"Hurricane: Writing Blog Posts with GFL","bloglink2":"https://weaviate.io/blog/multimodal-rag","blogTitle2":"Multimodal (RAG)","doclink":"https://docs.weaviate.io/weaviate/search/generative","docTitle":"Generative search","tags":["LLMs","Weaviate","Generative AI"],"cardImage":"generative-ai.jpg","next":"/learn/knowledgecards/large-language-model-llm","previous":"/learn/knowledgecards/chunking"},{"id":"llms-7","type":"LLMS","category":"Large Language Models","title":"Transformer Model","photo":"transformer-model-card-icon.svg","text":"A transformer model is a type of neural network architecture designed to handle sequential data, like text, for tasks such...","longText":"A transformer model is a type of neural network architecture designed to handle sequential data, like text, for tasks such as translation and question-answering. It\'s efficient and scalable, using attention mechanisms to weigh the importance of different parts of the input data.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/vector-embeddings-explained#transformer-models-bert-elmo-and-others","blogTitle":"Transformer models (BERT, ELMo, and others)","doclink":"https://docs.weaviate.io/weaviate/modules/qna-transformers#custom-qa-transformer-module","docTitle":"Custom Q&A Transformer module","tags":["LLMs","Weaviate","Transformer Model"],"cardImage":"transformer-model.jpg","next":"/learn/knowledgecards/large-language-model-llm","previous":"/learn/knowledgecards/generative-ai"},{"id":"irs-1","type":"IRS","category":"Information Retrieval/Search","title":"Reranking","photo":"reranking-card-icon.svg","text":"Re-ranking is adjusting the scores of search results after initial retrieval. It usually uses a machine learning model...","longText":"Re-ranking is adjusting the scores of search results after initial retrieval. It usually uses a machine learning model to reassess the relevance between the query and each result, refining the scores to improve accuracy and relevance for the user or an LLM\'s task. It\'s like a second quality check.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/ranking-models-for-better-search","blogTitle":"Ranking Models for Better Search","bloglink2":"https://weaviate.io/blog/llms-and-search#llms-in-re-ranking","blogTitle2":"LLMs in Re-Ranking","doclink":"https://docs.weaviate.io/weaviate/concepts/reranking","docT
1itle":"Reranking","tags":["LLMs","Weaviate","Reranking"],"cardImage":"reranking.jpg","next":"/learn/knowledgecards/retrieval-augmented-generation-rag","previous":"/learn/knowledgecards/retrieval-augmented-generation-rag"},{"id":"irs-2","type":"IRS","category":"Information Retrieval/Search","title":"Retrieval Augmented Generation (RAG)","photo":"retrieval-augmented-generation-card-icon.svg","text":"Retrieval Augmented Generation (RAG) is the process of contextualising prompts for large language models...","longText":"Retrieval Augmented Generation (RAG) is the process of contextualising prompts for large language models (LLMs) by leveraging available data in a store allowing an LLMs to ingest relevant information that guides their predictions.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/rag-evaluation#rag-metrics","blogTitle":"RAG Metrics","bloglink2":"https://weaviate.io/blog/multimodal-rag","blogTitle2":"Multimodal RAG","doclink":"https://docs.weaviate.io/weaviate/search/generativeg","docTitle":"Generative search","tags":["LLMs","Weaviate","rag"],"cardImage":"retrieval-augmented-generation.jpg","next":"/learn/knowledgecards/reranking","previous":"/learn/knowledgecards/reranking"},{"id":"et-1","type":"ET","category":"Embedding Types","title":"Variable Dimensions","photo":"variable-dimensions-card-icon.svg","text":"Flexible embedding sizes, like Matryoshka embeddings.\\nEncode information hierarchically, allowing adaptation...","longText":"Flexible embedding sizes, like Matryoshka embeddings. Encode information hierarchically, allowing adaptation to different tasks or computational constraints while preserving semantic meaning.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/openais-matryoshka-embeddings-in-weaviate","blogTitle":"OpenAI\'s Matryoshka Embeddings in Weaviate","videolink":"https://www.youtube.com/watch?v=ZvnKlUtMOkQ","videoTitle":"Matryoshka Representation Learning (MRL) for ML..","tags":["Embedding Types","Weaviate","Dimensions"],"cardImage":"variable-dimensions.jpg","next":"/learn/knowledgecards/sparse-embeddings","previous":"/learn/knowledgecards/binary-embeddings"},{"id":"et-2","type":"ET","category":"Embedding Types","title":"Sparse Embeddings","photo":"sparse-embeddings-card-icon.svg","text":"Sparse vectors are often high-dimensional with many zero values. They are generated from algorithms like BM25 and SPLADE...","longText":"Sparse vectors are often high-dimensional with many zero values. They are generated from algorithms like BM25 and SPLADE and are used in keyword-based search.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/hybrid-search-explained#bm25","blogTitle":"BM25","doclink":"https://docs.weaviate.io/weaviate/search/bm25","docTitle":"Search - BM25","tags":["Embedding Types","Weaviate","Sparse"],"cardImage":"sparse-embeddings.jpg","next":"/learn/knowledgecards/quantized-embeddings","previous":"/learn/knowledgecards/variable-dimensions"},{"id":"et-3","type":"ET","category":"Embedding Types","title":"Quantized Embeddings","photo":"quantized-embeddings-card-icon.svg","text":"Compressed dense vectors using lower-precision data types (e.g., float32 to int8). Reduces memory usage and speeds up search...","longText":"Compressed dense vectors using lower-precision data types (e.g., float32 to int8). Reduces memory usage and speeds up search while maintaining most semantic information.","link":"https://docs.weaviate.io/weaviate/quickstart","videolink":"https://www.youtube.com/watch?v=0diVrgyQwXA","videoTitle":"Vector Quantization Techniques with Etienne","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-quantization","docTitle":"Vector Quantization","tags":["Embedding Types","Weaviate","Quantized"],"cardImage":"sparse-embeddings.jpg","next":"/learn/knowledgecards/multivector-embeddings","previous":"/learn/knowledgecards/sparse-embeddings"},{"id":"et-4","type":"ET","category":"Embedding Types","title":"Multi-vector Embeddings","photo":"multi-vector-embeddings-card-icon.svg","text":"Usage of multiple vectors instead of one pooled vector to represent e.g., token-wise embeddings (e.g., ColBERT)...","longText":"Usage of multiple vectors instead of one pooled vector to represent e.g., token-wise embeddings (e.g., ColBERT). Allows for more detailed representation of complex texts.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://github.com/weaviate/recipes/blob/main/weaviate-features/named-vectors/NamedVectors-ColPali-POC.ipynb","docT
1itle":"How to use ColPali with Weaviate\'s Named Vectors!","tags":["Embedding Types","Weaviate","Multi-vector"],"cardImage":"multi-vector-embeddings.jpg","next":"/learn/knowledgecards/dense-embeddings","previous":"/learn/knowledgecards/quantized-embeddings"},{"id":"et-5","type":"ET","category":"Embedding Types","title":"Dense Embeddings","photo":"dense-embeddings-card-icon.svg","text":"Dense embeddings contain mostly non-zero values and are generated from machine learning models like Transformers...","longText":"Dense embeddings contain mostly non-zero values and are generated from machine learning models like Transformers. These vectors capture the semantic meaning of text and are used in semantic search.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/search/similarity","docTitle":"Similarity Search","tags":["Embedding Types","Weaviate","Dense"],"cardImage":"dense-embeddings.jpg","next":"/learn/knowledgecards/binary-embeddings","previous":"/learn/knowledgecards/multivector-embeddings"},{"id":"et-6","type":"ET","category":"Embedding Types","title":"Binary Embeddings","photo":"binary-embeddings-card-icon.svg","text":"Extreme quantization, reducing vector components to binary (0 or 1) values. Drastically reduces memory use...","longText":"Extreme quantization, reducing vector components to binary (0 or 1) values. Drastically reduces memory use.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/binary-quantization","blogTitle":"Binary Quantization","doclink":"https://docs.weaviate.io/weaviate/concepts/vector-quantization#binary-quantization","docTitle":"Vector Quantization","videolink":"https://www.youtube.com/watch?v=0diVrgyQwXA&t=2480s","videoTitle":"Vector Quantization Techniques with Etienne","tags":["Embedding Types","Weaviate","Binary"],"cardImage":"binary-embeddings.jpg","next":"/learn/knowledgecards/variable-dimensions","previous":"/learn/knowledgecards/dense-embeddings"},{"id":"ct-1","type":"CT","category":"Chunking Techniques","title":"Semantic Chunking","photo":"semantic-chunking-card-icon.svg","text":"In this technique, the text is divided into meaningful units, such as sentences or paragraphs, which are then vectorized...","longText":"In this technique, the text is divided into meaningful units, such as sentences or paragraphs, which are then vectorized. These units are then combined into chunks based on the cosine distance between their embeddings, with a new chunk formed whenever a significant context shift is detected.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/advanced-rag#2-semantic-chunking","blogTitle":"Advanced RAG","videolink":"https://www.youtube.com/live/LuhBgmwQeqw?si=OrVdanebGwhyaWHm&t=1769","videoTitle":"Dive into Chunking Strategies for RAG with Zain","tags":["Chunking Techniques","Weaviate","Semantic"],"cardImage":"semantic-chunking.jpg","next":"/learn/knowledgecards/recursive-chunking","previous":"/learn/knowledgecards/documentbased-chunking"},{"id":"ct-2","type":"CT","category":"Chunking Techniques","title":"Recursive Chunking","photo":"recursive-chunking-card-icon.svg","text":"Text is initially split using a primary separator, like paragraphs. If the resulting chunks are too large, secondary separators...","longText":"Text is initially split using a primary separator, like paragraphs. If the resulting chunks are too large, secondary separators, like sentences, are applied recursively until the desired chunk size is achieved. This technique respects the document\'s structure and is flexible for various use cases.","link":"https://docs.weaviate.io/weaviate/quickstart","videolink":"https://www.youtube.com/live/LuhBgmwQeqw?si=G84bHSHPczWX_mnS&t=1575","videoTitle":"Dive into Chunking Strategies for RAG with Zain","tags":["Chunking Techniques","Weaviate","Semantic"],"cardImage":"recursive-chunking.jpg","next":"/learn/knowledgecards/llmbased-chunking","previous":"/learn/knowledgecards/semantic-chunking"},{"id":"ct-3","type":"CT","category":"Chunking Techniques","title":"LLM-Based Chunking","photo":"llmbased-chunking-card-icon.svg","text":"This advanced technique uses a Language Model (LLM) to generate chunks. The LLM processes the text and generates semantically isolated sentences...","longText":"This advanced technique uses a Language Model (LLM) to generate chunks. The LLM processes the text and generates semantically isolated sentences or propositions that can stand alone. While this method is highly accurate, it is also the most computationally demanding.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/advanced-rag#3-language-model-based-chunking","blogTitle":"Advanced RAG","videolink":"https://www.youtube.com/live/LuhBgmwQeqw?si=dKmndf2SBNYwVWeo&t=2052","videoTitle":"Dive into Chunking Strategies for RAG with Zain","tags":["Chunking Techniques","Weaviate","LLM-Based"],"cardImage":"llmbased-chunking.jpg","next":"/learn/knowledgecards/fixed-size-chunking","previous":"/learn/knowledgecards/recursive-chunking"},{"id":"ct-4","type":"CT","category":"Chunking Techniques","title":"Fixed Size Chunking","photo":"fixed-size-chunking-card-icon.svg","text":"This technique splits the text into chunks of a fixed size, without considering natural breaks or the structure of the content...","longText":"This technique splits the text into chunks of a fixed size, without considering natural breaks or the structure of the content. It\'s simple and cost-effective, but lacks contextual awareness. To improve this, overlapping chunks can be used, allowing adjacent chunks to share some content.","link":"https://docs.weaviate.io/weaviate/quickstart","videolink":"https://www.youtube.com/live/LuhBgmwQeqw?si=dKmndf2SBNYwVWeo&t=2052","videoTitle":"Dive into Chunking Strategies for RAG with Zain","tags":["Chunking Techniques","Weaviate","Fixed Size"],"cardImage":"fixed-size-chunking.jpg","next":"/learn/knowledgecards/documentbased-chunking","previous":"/learn/knowledgecards/llmbased-chunking"},{"id":"ct-5","type":"CT","category":"Chunking Techniques","title":"Document-Based Chunking","photo":"documentbased-chunking-card-icon.svg","text":"This technique creates chunks based on the natural divisions within the document, such as headings or sections. It\'s very effective for structured data like HTML...","longText":"This technique creates chunks based on the natural divisions within the document, such as headings or sections. It\'s very effective for structured data like HTML, Markdown, or code files but it\u2019s less useful when the data lacks clear structural elements.","link":"https://docs.weaviate.io/weaviate/quickstart","videolink":"https://www.youtube.com/live/LuhBgmwQeqw?si=NOKa7LdoVDA_akLC&t=1685","videoTitle":"Dive into Chunking Strategies for RAG with Zain","tags":["Chunking Techniques","Weaviate","Fixed Size"],"cardImage":"documentbased-chunking.jpg","next":"/learn/knowledgecards/semantic-chunking","previous":"/learn/knowledgecards/fixed-size-chunking"},{"id":"art-1","type":"ART","category":"Advanced RAG Techniques","title":"Reasoning and Acting (ReAct)","photo":"react-card-icon.svg","text":"ReAct prompting combines CoT with agents, creating a system where the model can generate thoughts and delegate actions to agents that interact with external data sources...","longText":"ReAct prompting combines CoT with agents, creating a system where the model can generate thoughts and delegate actions to agents that interact with external data sources. ReAct enables LLMs to dynamically interact with retrieved documents, updating reasoning and actions based on external knowledge to provide more contextually relevant responses.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"reasoning-and-acting.jpg","next":"/learn/knowledgecards/tree-of-thoughts-tot","previous":"/learn/knowledgecards/query-expansion"},{"id":"art-2","type":"ART","category":"Advanced RAG Techniques","title":"Tree of Thoughts (ToT)","photo":"tree-of-thoughts-card-icon.svg","text":"Tree of Thoughts prompting builds on CoT by instructing the model to evaluate its responses at each step in the problem-solving process or even generate several different solutions to a problem...","longText":"Tree of Thoughts prompting builds on CoT by instructing the model to evaluate its responses at each step in the problem-solving process or even generate several different solutions to a problem and choose the best result. This is useful in RAG when there are many potential pieces of evidence, and the model needs to weigh different possible answers based on multiple retrieved documents.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"tree-of-thoughts.jpg","next":"/learn/knowledgecards/chain-of-thoughts-cot","previous":"/learn/knowledgecards/reasoning-and-acting-react"},{"id":"art-3","type":"ART","category":"Advanced RAG Techniques","title":"Chain of Thought (CoT)","slug":"/learn/knowledgecards/chain-of-thoughts-cot","photo":"chain-of-thought-card-icon.svg","text":"Chain of Thought (CoT) prompting involves asking the model to \u201cthink step-by-step\u201d and break down complex reasoning tasks into a series of intermediate steps...","longText":"Chain of Thought (CoT) prompting involves asking the model to \u201cthink step-by-step\u201d and break down complex reasoning tasks into a series of intermediate steps. This can be especially useful when retrieved documents contain conflicting or dense information that requires careful analysis.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"chain-of-thought.jpg","next":"/learn/knowledgecards/data-cleansing","previous":"/learn/knowledgecards/tree-of-thoughts-tot"},{"id":"art-4","type":"ART","category":"Advanced RAG Techniques","title":"Data Cleansing","photo":"data-cleansing-card-icon.svg","text":"Data cleaning and noise reduction involves removing irrelevant information (such as headers, footers, or boilerplate text), correcting inconsistencies, and handling missing values...","longText":"Data cleaning and noise reduction involves removing irrelevant information (such as headers, footers, or boilerplate text), correcting inconsistencies, and handling missing values while maintaining the extracted data\'s structural integrity.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"data-cleansing.jpg","next":"/learn/knowledgecards/data-extraction-and-parsing","previous":"/learn/knowledgecards/chain-of-thoughts-cot"},{"id":"art-5","type":"ART","category":"Advanced RAG Techniques","title":"Data Extraction and Parsing","photo":"data-extraction-and-parsing-card-icon.svg","text":"Data extraction and parsing convert raw data into an LLM-ready format. Text-based files retain structure, while OCR handles scanned docs...","longText":"Data extraction and parsing convert raw data into an LLM-ready format. Text-based files retain structure, while OCR handles scanned docs. Advanced multimodal models may replace OCR by embedding images directly. HTML parsing, spreadsheet handling, and metadata extraction refine the process for efficiency.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"data-extraction-and-parsing.jpg","next":"/learn/knowledgecards/data-transformation","previous":"/learn/knowledgecards/data-cleansing"},{"id":"art-6","type":"ART","category":"Advanced RAG Techniques","title":"Data Transformation","photo":"data-transformation-card-icon.svg","text":"Data Transformation involves convert
1ing all processed content into a standardized schema regardless of original file type...","longText":"Data Transformation involves converting all processed content into a standardized schema regardless of original file type. It is at this stage that document partitioning occurs, separating document content into logical units or elements (e.g., paragraphs, sections, tables).","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"data-transformation.jpg","next":"/learn/knowledgecards/embedding-model-fine-tuning","previous":"/learn/knowledgecards/data-extraction-and-parsing"},{"id":"art-7","type":"ART","category":"Advanced RAG Techniques","title":"Embedding Model Fine-Tuning","photo":"embedding-modal-fine-tuning-card-icon.svg","text":"Embedding model fine-tuning can significantly improve the quality of embeddings, subsequently improving performance on downstream tasks like RAG...","longText":"Embedding model fine-tuning can significantly improve the quality of embeddings, subsequently improving performance on downstream tasks like RAG. Fine-tuning improves embeddings to better capture the dataset\'s meaning and context, leading to more accurate and relevant retrievals in RAG applications.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","bloglink":"https://weaviate.io/blog/how-to-choose-an-embedding-model","blogTitle":"How to choose an embedding model","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"embedding-modal-fine-tuning.jpg","next":"/learn/knowledgecards/distance-thresholding","previous":"/learn/knowledgecards/data-transformation"},{"id":"art-8","type":"ART","category":"Advanced RAG Techniques","title":"Distance Thresholding","photo":"distance-thresholding-card-icon.svg","text":"Distance thresholding adds a quality check by setting a maximum allowed distance between vectors...","longText":"Distance thresholding adds a quality check by setting a maximum allowed distance between vectors. Any result with a distance score above this threshold gets filtered out, even if it would have made the top_k cutoff. This helps remove the obvious bad matches but requires careful threshold adjustment.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"distance-thresholding.jpg","next":"/learn/knowledgecards/prompt-engineering","previous":"/learn/knowledgecards/embedding-model-fine-tuning"},{"id":"art-9","type":"ART","category":"Advanced RAG Techniques","title":"Prompt Engineering","photo":"prompt-engineering-card-icon.svg","text":"Prompt engineering is the practice of optimizing LLM prompts to improve the quality and accuracy of generated output...","longText":"Prompt engineering is the practice of optimizing LLM prompts to improve the quality and accuracy of generated output. This process does not require making changes to the LLM itself making it an efficient and accessible way to enhance performance without complex modifications.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","bloglink":"https://weaviate.io/blog/dspy-optimizers","blogTitle":"Dspy Optimizers","videolink":"https://www.youtube.com/watch?v=skMH3DOV_UQ","videoTitle":"MIPRO and DSPy with Krista Opsahl-Ong!","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"prompt-engineering.jpg","next":"/learn/knowledgecards/context-compression","previous":"/learn/knowledgecards/distance-thresholding"},{"id":"art-10","type":"ART","category":"Advanced RAG Techniques","title":"Context Compression","photo":"context-compression-card-icon.svg","text":"Context compression refines RAG retrieval by filtering out irrelevant or redundant data, reducing token usage and costs...","longText":"Context compression refines RAG retrieval by filtering out irrelevant or redundant data, reducing token usage and costs. It uses a base retriever to find documents, and then a compressor to extract key info. Approaches like embedding-based and lexical-based compression preserve essential content while optimizing efficiency.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"context-compression.jpg","next":"/learn/knowledgecards/metadata-filtering","previous":"/learn/knowledgecards/prompt-engineering"},{"id":"art-11","type":"ART","category":"Advanced RAG Techniques","title":"Metadata Filtering","photo":"metadata-filtering-card-icon.svg","text":"Metadata is contextual information attached to documents in vector databases, including timestamps, categories, and source references...","longText":"Metadata is contextual information attached to documents in vector databases, including timestamps, categories, and source references. When combined with vector search capabilities, metadata enables precise filtering of results beyond pure semantic similarity, ensuring more accurate and relevant search outcomes.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://docs.weaviate.io/weaviate/search/filters","docTitle":"Filters","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"metadata-filtering.jpg","next":"/learn/knowledgecards/query-routing","previous":"/learn/knowledgecards/context-compression"},{"id":"art-12","type":"ART","category":"Advanced RAG Techniques","title":"Query Routing","photo":"query-routing-card-icon.svg","text":"Query routing directs queries to specific pipelines based on content and intent, optimizing RAG systems...","longText":"Query routing directs queries to specific pipelines based on content and intent, optimizing RAG systems. It uses multi-index strategies and agentic elements to determine the best retrieval method. This allows specialized handling, from fact-based queries to complex summarization, using diverse data stores and retrieval strategies.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","doclink2":"https://docs.weaviate.io/weaviate/tutorials/query","docTitle2":"Queries in detail","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"query-routing.jpg","next":"/learn/knowledgecards/data-pre-processing","previous":"/learn/knowledgecards/metadata-filtering"},{"id":"art-13","type":"ART","category":"Advanced RAG Techniques","title":"Data Pre-Processing","photo":"data-pre-processing-card-icon.svg","text":"Data pre-processing thoughtfully transforms your raw data into a structured format suitable for LLMs...","longText":"Data pre-processing thoughtfully transforms your raw data into a structured format suitable for LLMs. The optimal pre-processing techniques should be tailored to your specific use case or requirements. This process usually begins with data acquisition and integration, where diverse document types from multiple sources are collected and consolidated into a \\"knowledge base\\".","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"data-pre-processing.jpg","next":"/learn/knowledgecards/autocut","previous":"/learn/knowledgecards/query-routing"},{"id":"art-14","type":"ART","category":"Advanced RAG Techniques","title":"Autocut","photo":"autocut-card-icon.svg","text":"Autocut is a method to filter out irrelevant information retrieved from the database...","longText":"Autocut is a method to filter out irrelevant information retrieved from the database, which can otherwise mislead the LLM and cause hallucinations. It identifies a cutoff point where similarity scores drop significantly, excluding less relevant objects to ensure only the most pertinent information is used.","link":"https://docs.weaviate.io/weaviate/quickstart","bloglink":"https://weaviate.io/blog/advanced-rag#1-autocut-to-remove-irrelevant-information","blogTitle":"Advanced RAG","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"autocut.jpg","next":"/learn/knowledgecards/query-rewriting","previous":"/learn/knowledgecards/data-pre-processing"},{"id":"art-15","type":"ART","category":"Advanced RAG Techniques","title":"Query Rewriting","photo":"query-rewriting-card-icon.svg","text":"Query Rewriting involves reformulating the original user query to make it more suitable for retrieval...","longText":"Query Rewriting involves reformulating the original user query to make it more suitable for retrieval. This is particularly useful in scenarios where user queries are not optimally phrased or expressed differently. This can be achieved by using an LLM to rephrase the original user query or employing specialized smaller language models trained specifically for this task.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"query-rewriting.jpg","next":"/learn/knowledgecards/query-expansion","previous":"/learn/knowledgecards/autocut"},{"id":"art-15","type":"ART","category":"Advanced RAG Techniques","title":"Query Expansion","photo":"query-expansion-card-icon.svg","text":"Query Expansion focuses on broadening the original query to capture more relevant information.","longText":"Query Expansion focuses on broadening the original query to capture more relevant information. This involves using an LLM to generate multiple similar queries based on the user\'s initial input. The
1se expanded queries are then used in the retrieval process, increasing both the number and relevance of retrieved documents.","link":"https://docs.weaviate.io/weaviate/quickstart","doclink":"https://weaviate.io/ebooks/advanced-rag-techniques","docTitle":"Advanced RAG Techniques","tags":["RAG","Weaviate","Advanced RAG Techniques"],"cardImage":"query-expansion.jpg","next":"/learn/knowledgecards/reasoning-and-acting-react","previous":"/learn/knowledgecards/query-rewriting"}]}')}}]);
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.