From 1ffa56897446cb36bd1e0a5699b6bc9c89e844a4 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 09:08:53 +0530 Subject: [PATCH 01/14] Addition of Pinecone MMR search --- .../nodes/vectorstores/Pinecone/Pinecone.ts | 63 +++++++++++++++++-- 1 file changed, 59 insertions(+), 4 deletions(-) diff --git a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts index 4e6967bcf..3f2e3ef1f 100644 --- a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts +++ b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts @@ -23,11 +23,11 @@ class Pinecone_VectorStores implements INode { constructor() { this.label = 'Pinecone' this.name = 'pinecone' - this.version = 1.0 + this.version = 2.0 this.type = 'Pinecone' this.icon = 'pinecone.svg' this.category = 'Vector Stores' - this.description = `Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database` + this.description = `Upsert embedded data and perform search upon query using Pinecone, a leading fully managed hosted vector database` this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -77,6 +77,43 @@ class Pinecone_VectorStores implements INode { type: 'number', additionalParams: true, optional: true + }, + { + label: 'Search Type', + name: 'searchType', + type: 'options', + default: 'similarity', + options: [ + { + label: 'Similarity', + name: 'similarity' + }, + { + label: 'Max Marginal Relevance', + name: 'mmr' + } + ], + additionalParams: true, + optional: true + }, + { + label: 'Fetch K (for MMR Search)', + name: 'fetchK', + description: 'Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR', + placeholder: '20', + type: 'number', + additionalParams: true, + optional: true + }, + { + label: 'Lambda (for MMR Search)', + name: 'lambda', + description: + 'Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR', + placeholder: '0.5', + type: 'number', + additionalParams: true, + optional: true } ] this.outputs = [ @@ -141,6 +178,7 @@ class Pinecone_VectorStores implements INode { const docs = nodeData.inputs?.document as Document[] const embeddings = nodeData.inputs?.embeddings as Embeddings const output = nodeData.outputs?.output as string + const searchType = nodeData.outputs?.searchType as string const topK = nodeData.inputs?.topK as string const k = topK ? parseFloat(topK) : 4 @@ -176,8 +214,25 @@ class Pinecone_VectorStores implements INode { const vectorStore = await PineconeStore.fromExistingIndex(embeddings, obj) if (output === 'retriever') { - const retriever = vectorStore.asRetriever(k) - return retriever + if ('mmr' === searchType) { + const fetchK = nodeData.inputs?.fetchK as string + const lambda = nodeData.inputs?.lambda as string + const f = fetchK ? parseInt(fetchK) : 20 + const l = lambda ? parseFloat(lambda) : 0.5 + const retriever = vectorStore.asRetriever({ + searchType: 'mmr', + k: 5, + searchKwargs: { + fetchK: f, + lambda: l + } + }) + return retriever + } else { + // "searchType" is "similarity" + const retriever = vectorStore.asRetriever(k) + return retriever + } } else if (output === 'vectorStore') { ;(vectorStore as any).k = k return vectorStore From 579f66e57e3da71e0a53db10cf1baeb986d4bb31 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 09:20:55 +0530 Subject: [PATCH 02/14] Updating langchain to 0.0.198 --- packages/components/package.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/packages/components/package.json b/packages/components/package.json index 9cb0bf1e9..018d7a77e 100644 --- a/packages/components/package.json +++ b/packages/components/package.json @@ -52,7 +52,7 @@ "html-to-text": "^9.0.5", "husky": "^8.0.3", "ioredis": "^5.3.2", - "langchain": "^0.0.196", + "langchain": "^0.0.198", "langfuse": "^1.2.0", "langfuse-langchain": "^1.0.31", "langsmith": "^0.0.49", From 56b043264a5370ac041f8fde1adf928b5068eace Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:08:48 +0530 Subject: [PATCH 03/14] MMR Search for Pinecone, Weaviate, Zep and Supabase --- .../nodes/vectorstores/Pinecone/Pinecone.ts | 73 ++---------------- .../nodes/vectorstores/Supabase/Supabase.ts | 18 ++--- .../nodes/vectorstores/VectorStoreUtils.ts | 76 +++++++++++++++++++ .../nodes/vectorstores/Weaviate/Weaviate.ts | 18 ++--- .../components/nodes/vectorstores/Zep/Zep.ts | 22 ++---- 5 files changed, 98 insertions(+), 109 deletions(-) create mode 100644 packages/components/nodes/vectorstores/VectorStoreUtils.ts diff --git a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts index 3f2e3ef1f..4b91a9b54 100644 --- a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts +++ b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts @@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base' import { Document } from 'langchain/document' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class Pinecone_VectorStores implements INode { label: string @@ -23,11 +24,11 @@ class Pinecone_VectorStores implements INode { constructor() { this.label = 'Pinecone' this.name = 'pinecone' - this.version = 2.0 + this.version = 3.0 this.type = 'Pinecone' this.icon = 'pinecone.svg' this.category = 'Vector Stores' - this.description = `Upsert embedded data and perform search upon query using Pinecone, a leading fully managed hosted vector database` + this.description = `Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database` this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -77,45 +78,9 @@ class Pinecone_VectorStores implements INode { type: 'number', additionalParams: true, optional: true - }, - { - label: 'Search Type', - name: 'searchType', - type: 'options', - default: 'similarity', - options: [ - { - label: 'Similarity', - name: 'similarity' - }, - { - label: 'Max Marginal Relevance', - name: 'mmr' - } - ], - additionalParams: true, - optional: true - }, - { - label: 'Fetch K (for MMR Search)', - name: 'fetchK', - description: 'Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR', - placeholder: '20', - type: 'number', - additionalParams: true, - optional: true - }, - { - label: 'Lambda (for MMR Search)', - name: 'lambda', - description: - 'Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR', - placeholder: '0.5', - type: 'number', - additionalParams: true, - optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'Pinecone Retriever', @@ -177,10 +142,6 @@ class Pinecone_VectorStores implements INode { const pineconeMetadataFilter = nodeData.inputs?.pineconeMetadataFilter const docs = nodeData.inputs?.document as Document[] const embeddings = nodeData.inputs?.embeddings as Embeddings - const output = nodeData.outputs?.output as string - const searchType = nodeData.outputs?.searchType as string - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 const credentialData = await getCredentialData(nodeData.credential ?? '', options) const pineconeApiKey = getCredentialParam('pineconeApiKey', credentialData, nodeData) @@ -213,31 +174,7 @@ class Pinecone_VectorStores implements INode { const vectorStore = await PineconeStore.fromExistingIndex(embeddings, obj) - if (output === 'retriever') { - if ('mmr' === searchType) { - const fetchK = nodeData.inputs?.fetchK as string - const lambda = nodeData.inputs?.lambda as string - const f = fetchK ? parseInt(fetchK) : 20 - const l = lambda ? parseFloat(lambda) : 0.5 - const retriever = vectorStore.asRetriever({ - searchType: 'mmr', - k: 5, - searchKwargs: { - fetchK: f, - lambda: l - } - }) - return retriever - } else { - // "searchType" is "similarity" - const retriever = vectorStore.asRetriever(k) - return retriever - } - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } diff --git a/packages/components/nodes/vectorstores/Supabase/Supabase.ts b/packages/components/nodes/vectorstores/Supabase/Supabase.ts index 13840ab78..a5477914f 100644 --- a/packages/components/nodes/vectorstores/Supabase/Supabase.ts +++ b/packages/components/nodes/vectorstores/Supabase/Supabase.ts @@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils' import { SupabaseLibArgs, SupabaseVectorStore } from 'langchain/vectorstores/supabase' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class Supabase_VectorStores implements INode { label: string @@ -23,11 +24,11 @@ class Supabase_VectorStores implements INode { constructor() { this.label = 'Supabase' this.name = 'supabase' - this.version = 1.0 + this.version = 2.0 this.type = 'Supabase' this.icon = 'supabase.svg' this.category = 'Vector Stores' - this.description = 'Upsert embedded data and perform similarity search upon query using Supabase via pgvector extension' + this.description = 'Upsert embedded data and perform similarity or mmr search upon query using Supabase via pgvector extension' this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -81,6 +82,7 @@ class Supabase_VectorStores implements INode { optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'Supabase Retriever', @@ -135,9 +137,6 @@ class Supabase_VectorStores implements INode { const queryName = nodeData.inputs?.queryName as string const embeddings = nodeData.inputs?.embeddings as Embeddings const supabaseMetadataFilter = nodeData.inputs?.supabaseMetadataFilter - const output = nodeData.outputs?.output as string - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 const credentialData = await getCredentialData(nodeData.credential ?? '', options) const supabaseApiKey = getCredentialParam('supabaseApiKey', credentialData, nodeData) @@ -157,14 +156,7 @@ class Supabase_VectorStores implements INode { const vectorStore = await SupabaseVectorStore.fromExistingIndex(embeddings, obj) - if (output === 'retriever') { - const retriever = vectorStore.asRetriever(k) - return retriever - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } diff --git a/packages/components/nodes/vectorstores/VectorStoreUtils.ts b/packages/components/nodes/vectorstores/VectorStoreUtils.ts new file mode 100644 index 000000000..b63a4121c --- /dev/null +++ b/packages/components/nodes/vectorstores/VectorStoreUtils.ts @@ -0,0 +1,76 @@ +import { INodeData } from '../../src' + +export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: any) => { + const output = nodeData.outputs?.output as string + const searchType = nodeData.outputs?.searchType as string + const topK = nodeData.inputs?.topK as string + const k = topK ? parseFloat(topK) : 4 + + if (output === 'retriever') { + if ('mmr' === searchType) { + const fetchK = nodeData.inputs?.fetchK as string + const lambda = nodeData.inputs?.lambda as string + const f = fetchK ? parseInt(fetchK) : 20 + const l = lambda ? parseFloat(lambda) : 0.5 + const retriever = vectorStore.asRetriever({ + searchType: 'mmr', + k: k, + searchKwargs: { + fetchK: f, + lambda: l + } + }) + return retriever + } else { + // "searchType" is "similarity" + return vectorStore.asRetriever(k) + } + } else if (output === 'vectorStore') { + ;(vectorStore as any).k = k + return vectorStore + } +} + +export const addMMRInputParams = (inputs: any[]) => { + const mmrInputParams = [ + { + label: 'Search Type', + name: 'searchType', + type: 'options', + default: 'similarity', + options: [ + { + label: 'Similarity', + name: 'similarity' + }, + { + label: 'Max Marginal Relevance', + name: 'mmr' + } + ], + additionalParams: true, + optional: true + }, + { + label: 'Fetch K (for MMR Search)', + name: 'fetchK', + description: 'Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR', + placeholder: '20', + type: 'number', + additionalParams: true, + optional: true + }, + { + label: 'Lambda (for MMR Search)', + name: 'lambda', + description: + 'Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR', + placeholder: '0.5', + type: 'number', + additionalParams: true, + optional: true + } + ] + + inputs.push(...mmrInputParams) +} diff --git a/packages/components/nodes/vectorstores/Weaviate/Weaviate.ts b/packages/components/nodes/vectorstores/Weaviate/Weaviate.ts index 5c31c7371..0eface588 100644 --- a/packages/components/nodes/vectorstores/Weaviate/Weaviate.ts +++ b/packages/components/nodes/vectorstores/Weaviate/Weaviate.ts @@ -5,6 +5,7 @@ import { Document } from 'langchain/document' import { Embeddings } from 'langchain/embeddings/base' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class Weaviate_VectorStores implements INode { label: string @@ -23,12 +24,12 @@ class Weaviate_VectorStores implements INode { constructor() { this.label = 'Weaviate' this.name = 'weaviate' - this.version = 1.0 + this.version = 2.0 this.type = 'Weaviate' this.icon = 'weaviate.png' this.category = 'Vector Stores' this.description = - 'Upsert embedded data and perform similarity search upon query using Weaviate, a scalable open-source vector database' + 'Upsert embedded data and perform similarity or mmr search using Weaviate, a scalable open-source vector database' this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -107,6 +108,7 @@ class Weaviate_VectorStores implements INode { optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'Weaviate Retriever', @@ -174,9 +176,6 @@ class Weaviate_VectorStores implements INode { const weaviateTextKey = nodeData.inputs?.weaviateTextKey as string const weaviateMetadataKeys = nodeData.inputs?.weaviateMetadataKeys as string const embeddings = nodeData.inputs?.embeddings as Embeddings - const output = nodeData.outputs?.output as string - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 const credentialData = await getCredentialData(nodeData.credential ?? '', options) const weaviateApiKey = getCredentialParam('weaviateApiKey', credentialData, nodeData) @@ -199,14 +198,7 @@ class Weaviate_VectorStores implements INode { const vectorStore = await WeaviateStore.fromExistingIndex(embeddings, obj) - if (output === 'retriever') { - const retriever = vectorStore.asRetriever(k) - return retriever - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } diff --git a/packages/components/nodes/vectorstores/Zep/Zep.ts b/packages/components/nodes/vectorstores/Zep/Zep.ts index ebb13c643..3d9f19787 100644 --- a/packages/components/nodes/vectorstores/Zep/Zep.ts +++ b/packages/components/nodes/vectorstores/Zep/Zep.ts @@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base' import { Document } from 'langchain/document' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class Zep_VectorStores implements INode { label: string @@ -23,12 +24,12 @@ class Zep_VectorStores implements INode { constructor() { this.label = 'Zep' this.name = 'zep' - this.version = 1.0 + this.version = 2.0 this.type = 'Zep' this.icon = 'zep.svg' this.category = 'Vector Stores' this.description = - 'Upsert embedded data and perform similarity search upon query using Zep, a fast and scalable building block for LLM apps' + 'Upsert embedded data and perform similarity or mmr search upon query using Zep, a fast and scalable building block for LLM apps' this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -88,6 +89,7 @@ class Zep_VectorStores implements INode { optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'Zep Retriever', @@ -144,9 +146,6 @@ class Zep_VectorStores implements INode { const zepMetadataFilter = nodeData.inputs?.zepMetadataFilter const dimension = nodeData.inputs?.dimension as number const embeddings = nodeData.inputs?.embeddings as Embeddings - const output = nodeData.outputs?.output as string - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 const credentialData = await getCredentialData(nodeData.credential ?? '', options) const apiKey = getCredentialParam('apiKey', credentialData, nodeData) @@ -165,14 +164,7 @@ class Zep_VectorStores implements INode { const vectorStore = await ZepExistingVS.fromExistingIndex(embeddings, zepConfig) - if (output === 'retriever') { - const retriever = vectorStore.asRetriever(k) - return retriever - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } @@ -210,7 +202,7 @@ class ZepExistingVS extends ZepVectorStore { this.args = args } - async initalizeCollection(args: IZepConfig & Partial) { + async initializeCollection(args: IZepConfig & Partial) { this.client = await ZepClient.init(args.apiUrl, args.apiKey) try { this.collection = await this.client.document.getCollection(args.collectionName) @@ -259,7 +251,7 @@ class ZepExistingVS extends ZepVectorStore { const newfilter = { where: { and: ANDFilters } } - await this.initalizeCollection(this.args!).catch((err) => { + await this.initializeCollection(this.args!).catch((err) => { console.error('Error initializing collection:', err) throw err }) From 94236c4b5fe80a3449eed29f36ddb9e620077abe Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:12:04 +0530 Subject: [PATCH 04/14] Compression Retriever - Cohere Rerank --- .../CohereRerankRetriever/CohereRerank.ts | 51 ++++++++++++ .../CohereRerankRetriever.ts | 77 +++++++++++++++++++ .../compressionRetriever.svg | 7 ++ 3 files changed, 135 insertions(+) create mode 100644 packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts create mode 100644 packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts create mode 100644 packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts new file mode 100644 index 000000000..612581edb --- /dev/null +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts @@ -0,0 +1,51 @@ +import { Callbacks } from 'langchain/callbacks' +import { Document } from 'langchain/document' +import { BaseDocumentCompressor } from 'langchain/retrievers/document_compressors' +import axios from 'axios' +export class CohereRerank extends BaseDocumentCompressor { + private cohereAPIKey: any + private COHERE_API_URL = 'https://api.cohere.ai/v1/rerank' + private model: string + + constructor(cohereAPIKey: string, model: string) { + super() + this.cohereAPIKey = cohereAPIKey + this.model = model + } + async compressDocuments( + documents: Document>[], + query: string, + _?: Callbacks | undefined + ): Promise>[]> { + // avoid empty api call + if (documents.length === 0) { + return [] + } + const config = { + headers: { + Authorization: `Bearer ${this.cohereAPIKey}`, + 'Content-Type': 'application/json', + Accept: 'application/json' + } + } + const data = { + model: this.model, + max_chunks_per_doc: 10, + query: query, + return_documents: false, + documents: documents.map((doc) => doc.pageContent) + } + try { + let returnedDocs = await axios.post(this.COHERE_API_URL, data, config) + const finalResults: Document>[] = [] + returnedDocs.data.results.forEach((result: any) => { + const doc = documents[result.index] + doc.metadata.relevance_score = result.relevance_score + finalResults.push(doc) + }) + return finalResults + } catch (error) { + return documents + } + } +} diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts new file mode 100644 index 000000000..2e7090bcf --- /dev/null +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts @@ -0,0 +1,77 @@ +import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' +import { BaseRetriever } from 'langchain/schema/retriever' +import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' +import { getCredentialData, getCredentialParam } from '../../../src' +import { CohereRerank } from './CohereRerank' + +class CohereRerankRetriever_Retrievers implements INode { + label: string + name: string + version: number + description: string + type: string + icon: string + category: string + baseClasses: string[] + inputs: INodeParams[] + outputs: INodeOutputsValue[] + credential: INodeParams + badge: string + + constructor() { + this.label = 'Cohere Rerank Retriever' + this.name = 'cohereRerankRetriever' + this.version = 1.0 + this.type = 'Cohere Rerank Retriever' + this.icon = 'compressionRetriever.svg' + this.category = 'Retrievers' + this.badge = 'NEW' + this.description = 'Cohere Rerank indexes the documents from most to least semantically relevant to the query.' + this.baseClasses = [this.type, 'BaseRetriever'] + this.credential = { + label: 'Connect Credential', + name: 'credential', + type: 'credential', + credentialNames: ['cohereApi'] + } + this.inputs = [ + { + label: 'Base Retriever', + name: 'baseRetriever', + type: 'VectorStoreRetriever' + }, + { + label: 'Model Name', + name: 'model', + type: 'options', + options: [ + { + label: 'rerank-english-v2.0', + name: 'rerank-english-v2.0' + }, + { + label: 'rerank-multilingual-v2.0', + name: 'rerank-multilingual-v2.0' + } + ], + default: 'rerank-english-v2.0', + optional: true + } + ] + } + + async init(nodeData: INodeData, _: string, options: ICommonObject): Promise { + const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever + const model = nodeData.inputs?.model as string + const credentialData = await getCredentialData(nodeData.credential ?? '', options) + const cohereApiKey = getCredentialParam('cohereApiKey', credentialData, nodeData) + + const cohereCompressor = new CohereRerank(cohereApiKey, model) + return new ContextualCompressionRetriever({ + baseCompressor: cohereCompressor, + baseRetriever: baseRetriever + }) + } +} + +module.exports = { nodeClass: CohereRerankRetriever_Retrievers } diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg new file mode 100644 index 000000000..23c52d25e --- /dev/null +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg @@ -0,0 +1,7 @@ + + + + + + + \ No newline at end of file From fb02632f4b682b8b0ee2d1ec1a79330118318eaa Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:12:16 +0530 Subject: [PATCH 05/14] Compression Retriever - Embeddings Filter --- .../EmbeddingsFilterRetriever.ts | 97 +++++++++++++++++++ .../compressionRetriever.svg | 7 ++ 2 files changed, 104 insertions(+) create mode 100644 packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts create mode 100644 packages/components/nodes/retrievers/EmbeddingsFilterRetriever/compressionRetriever.svg diff --git a/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts new file mode 100644 index 000000000..d373704c0 --- /dev/null +++ b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts @@ -0,0 +1,97 @@ +import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' +import { BaseRetriever } from 'langchain/schema/retriever' +import { Embeddings } from 'langchain/embeddings/base' +import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' +import { EmbeddingsFilter } from 'langchain/retrievers/document_compressors/embeddings_filter' + +class EmbeddingsFilterRetriever_Retrievers implements INode { + label: string + name: string + version: number + description: string + type: string + icon: string + category: string + baseClasses: string[] + inputs: INodeParams[] + outputs: INodeOutputsValue[] + badge: string + + constructor() { + this.label = 'Embeddings Filter Retriever' + this.name = 'embeddingsFilterRetriever' + this.version = 1.0 + this.type = 'EmbeddingsFilterRetriever' + this.icon = 'compressionRetriever.svg' + this.category = 'Retrievers' + this.badge = 'NEW' + this.description = 'A document compressor that uses embeddings to drop documents unrelated to the query' + this.baseClasses = [this.type, 'BaseRetriever'] + this.inputs = [ + { + label: 'Base Retriever', + name: 'baseRetriever', + type: 'VectorStoreRetriever' + }, + { + label: 'Embeddings', + name: 'embeddings', + type: 'Embeddings', + optional: false + }, + { + label: 'Similarity Threshold', + name: 'similarityThreshold', + description: + 'Threshold for determining when two documents are similar enough to be considered redundant. Must be specified if `k` is not set', + type: 'number', + default: 0.8, + step: 0.1, + optional: true + }, + { + label: 'K', + name: 'k', + description: + 'The number of relevant documents to return. Can be explicitly set to undefined, in which case similarity_threshold must be specified. Defaults to 20', + type: 'number', + default: 20, + step: 1, + optional: true, + additionalParams: true + } + ] + } + + async init(nodeData: INodeData): Promise { + const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever + const embeddings = nodeData.inputs?.embeddings as Embeddings + const similarityThreshold = nodeData.inputs?.similarityThreshold as string + const k = nodeData.inputs?.k as string + + if (k === undefined && similarityThreshold === undefined) { + throw new Error(`Must specify one of "k" or "similarity_threshold".`) + } + + let similarityThresholdNumber = 0.8 + if (similarityThreshold) { + similarityThresholdNumber = parseFloat(similarityThreshold) + } + let kNumber = 0.8 + if (k) { + kNumber = parseFloat(k) + } + const baseCompressor = new EmbeddingsFilter({ + embeddings: embeddings, + similarityThreshold: similarityThresholdNumber, + k: kNumber + }) + + return new ContextualCompressionRetriever({ + baseCompressor, + baseRetriever: baseRetriever + }) + } +} + +module.exports = { nodeClass: EmbeddingsFilterRetriever_Retrievers } diff --git a/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/compressionRetriever.svg new file mode 100644 index 000000000..23c52d25e --- /dev/null +++ b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/compressionRetriever.svg @@ -0,0 +1,7 @@ + + + + + + + \ No newline at end of file From f2f8ed6a9ce1ab41bd69193ce283757cd28cb16e Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:12:25 +0530 Subject: [PATCH 06/14] Compression Retriever - LLM filter --- .../LLMFilterCompressionRetriever.ts | 60 +++++++++++++++++++ .../compressionRetriever.svg | 7 +++ 2 files changed, 67 insertions(+) create mode 100644 packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts create mode 100644 packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts new file mode 100644 index 000000000..c421c7ce1 --- /dev/null +++ b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts @@ -0,0 +1,60 @@ +import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' +import { BaseRetriever } from 'langchain/schema/retriever' +import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' +import { BaseLanguageModel } from 'langchain/base_language' +import { LLMChainExtractor } from 'langchain/retrievers/document_compressors/chain_extract' + +class LLMFilterCompressionRetriever_Retrievers implements INode { + label: string + name: string + version: number + description: string + type: string + icon: string + category: string + baseClasses: string[] + inputs: INodeParams[] + outputs: INodeOutputsValue[] + badge: string + + constructor() { + this.label = 'LLM Filter Retriever' + this.name = 'llmFilterRetriever' + this.version = 1.0 + this.type = 'LLMFilterRetriever' + this.icon = 'compressionRetriever.svg' + this.category = 'Retrievers' + this.badge = 'NEW' + this.description = + 'Iterate over the initially returned documents and extract, from each, only the content that is relevant to the query' + this.baseClasses = [this.type, 'BaseRetriever'] + this.inputs = [ + { + label: 'Base Retriever', + name: 'baseRetriever', + type: 'VectorStoreRetriever' + }, + { + label: 'Language Model', + name: 'model', + type: 'BaseLanguageModel', + optional: true + }, + ] + } + + async init(nodeData: INodeData): Promise { + const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever + const model = nodeData.inputs?.model as BaseLanguageModel + + if (model) { + return new ContextualCompressionRetriever({ + baseCompressor: LLMChainExtractor.fromLLM(model), + baseRetriever: baseRetriever + }) + } + return {} + } +} + +module.exports = { nodeClass: LLMFilterCompressionRetriever_Retrievers } diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg new file mode 100644 index 000000000..23c52d25e --- /dev/null +++ b/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg @@ -0,0 +1,7 @@ + + + + + + + \ No newline at end of file From be79adb07c22586d4debd5384268463a874b878e Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:13:52 +0530 Subject: [PATCH 07/14] lint fixes --- .../LLMFilterRetriever/LLMFilterCompressionRetriever.ts | 2 +- packages/components/nodes/vectorstores/VectorStoreUtils.ts | 5 ++--- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts index c421c7ce1..e044468f4 100644 --- a/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts +++ b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts @@ -39,7 +39,7 @@ class LLMFilterCompressionRetriever_Retrievers implements INode { name: 'model', type: 'BaseLanguageModel', optional: true - }, + } ] } diff --git a/packages/components/nodes/vectorstores/VectorStoreUtils.ts b/packages/components/nodes/vectorstores/VectorStoreUtils.ts index b63a4121c..a01d43c4b 100644 --- a/packages/components/nodes/vectorstores/VectorStoreUtils.ts +++ b/packages/components/nodes/vectorstores/VectorStoreUtils.ts @@ -1,4 +1,4 @@ -import { INodeData } from '../../src' +import { INodeData } from "../../src"; export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: any) => { const output = nodeData.outputs?.output as string @@ -12,7 +12,7 @@ export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: const lambda = nodeData.inputs?.lambda as string const f = fetchK ? parseInt(fetchK) : 20 const l = lambda ? parseFloat(lambda) : 0.5 - const retriever = vectorStore.asRetriever({ + return vectorStore.asRetriever({ searchType: 'mmr', k: k, searchKwargs: { @@ -20,7 +20,6 @@ export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: lambda: l } }) - return retriever } else { // "searchType" is "similarity" return vectorStore.asRetriever(k) From f6ee137ca3e60b350f794181b43453eff520474b Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 22 Dec 2023 12:14:10 +0530 Subject: [PATCH 08/14] lint fixes --- packages/components/nodes/vectorstores/VectorStoreUtils.ts | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/packages/components/nodes/vectorstores/VectorStoreUtils.ts b/packages/components/nodes/vectorstores/VectorStoreUtils.ts index a01d43c4b..0d92587f4 100644 --- a/packages/components/nodes/vectorstores/VectorStoreUtils.ts +++ b/packages/components/nodes/vectorstores/VectorStoreUtils.ts @@ -1,4 +1,4 @@ -import { INodeData } from "../../src"; +import { INodeData } from '../../src' export const resolveVectorStoreOrRetriever = (nodeData: INodeData, vectorStore: any) => { const output = nodeData.outputs?.output as string From 4dd2f245ffdfb2fac5a169c97f9c825e7aa04828 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 29 Dec 2023 20:35:42 +0530 Subject: [PATCH 09/14] Compression Retriever: Reciprocal Rank Fusion --- .../retrievers/RRFRetriever/RRFRetriever.ts | 84 ++++++++++++++++ .../RRFRetriever/ReciprocalRankFusion.ts | 95 +++++++++++++++++++ .../RRFRetriever/compressionRetriever.svg | 7 ++ 3 files changed, 186 insertions(+) create mode 100644 packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts create mode 100644 packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts create mode 100644 packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg diff --git a/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts new file mode 100644 index 000000000..8d6d9d6fe --- /dev/null +++ b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts @@ -0,0 +1,84 @@ +import { INode, INodeData, INodeParams } from '../../../src/Interface' +import { BaseLanguageModel } from 'langchain/base_language' +import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' +import { BaseRetriever } from 'langchain/schema/retriever' +import { ReciprocalRankFusion } from './ReciprocalRankFusion' +import { VectorStoreRetriever } from 'langchain/vectorstores/base' + +class RRFRetriever_Retrievers implements INode { + label: string + name: string + version: number + description: string + type: string + icon: string + category: string + baseClasses: string[] + inputs: INodeParams[] + badge: string + + constructor() { + this.label = 'Reciprocal Rank Fusion Retriever' + this.name = 'RRFRetriever' + this.version = 2.0 + this.type = 'RRFRetriever' + this.badge = 'NEW' + this.icon = 'compressionRetriever.svg' + this.category = 'Retrievers' + this.description = 'Reciprocal Rank Fusion to re-rank search results by multiple query generation.' + this.baseClasses = [this.type, 'BaseRetriever'] + this.inputs = [ + { + label: 'Base Retriever', + name: 'baseRetriever', + type: 'VectorStoreRetriever' + }, + { + label: 'Language Model', + name: 'model', + type: 'BaseLanguageModel' + }, + { + label: 'Query Count', + name: 'queryCount', + description: 'Number of synthetic queries to generate. Default to 4', + placeholder: '4', + type: 'number', + default: 4, + additionalParams: true, + optional: true + }, + { + label: 'Top K', + name: 'topK', + description: 'Number of top results to fetch. Default to the TopK of the Base Retriever', + placeholder: '0', + type: 'number', + default: 0, + additionalParams: true, + optional: true + } + ] + } + + async init(nodeData: INodeData): Promise { + const llm = nodeData.inputs?.model as BaseLanguageModel + const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever + const queryCount = nodeData.inputs?.queryCount as string + const q = queryCount ? parseFloat(queryCount) : 4 + const topK = nodeData.inputs?.topK as string + let k = topK ? parseFloat(topK) : 4 + + if (k <= 0) { + k = (baseRetriever as VectorStoreRetriever).k + } + + const ragFusion = new ReciprocalRankFusion(llm, baseRetriever as VectorStoreRetriever, q, k) + return new ContextualCompressionRetriever({ + baseCompressor: ragFusion, + baseRetriever: baseRetriever + }) + } +} + +module.exports = { nodeClass: RRFRetriever_Retrievers } diff --git a/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts new file mode 100644 index 000000000..134d7c8a5 --- /dev/null +++ b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts @@ -0,0 +1,95 @@ +import { BaseDocumentCompressor } from 'langchain/retrievers/document_compressors' +import { Document } from 'langchain/document' +import { Callbacks } from 'langchain/callbacks' +import { BaseLanguageModel } from 'langchain/base_language' +import { ChatPromptTemplate, HumanMessagePromptTemplate, SystemMessagePromptTemplate } from 'langchain/prompts' +import { LLMChain } from 'langchain/chains' +import { VectorStoreRetriever } from 'langchain/vectorstores/base' + +export class ReciprocalRankFusion extends BaseDocumentCompressor { + private readonly llm: BaseLanguageModel + private readonly queryCount: number + private readonly topK: number + private baseRetriever: VectorStoreRetriever + constructor(llm: BaseLanguageModel, baseRetriever: VectorStoreRetriever, queryCount: number, topK: number) { + super() + this.queryCount = queryCount + this.llm = llm + this.baseRetriever = baseRetriever + this.topK = topK + } + async compressDocuments( + documents: Document>[], + query: string, + _?: Callbacks | undefined + ): Promise>[]> { + // avoid empty api call + if (documents.length === 0) { + return [] + } + const chatPrompt = ChatPromptTemplate.fromMessages([ + SystemMessagePromptTemplate.fromTemplate( + 'You are a helpful assistant that generates multiple search queries based on a single input query.' + ), + HumanMessagePromptTemplate.fromTemplate( + 'Generate multiple search queries related to: {input}. Provide these alternative questions separated by newlines, do not add any numbers.' + ), + HumanMessagePromptTemplate.fromTemplate('OUTPUT (' + this.queryCount + ' queries):') + ]) + const llmChain = new LLMChain({ + llm: this.llm, + prompt: chatPrompt + }) + const multipleQueries = await llmChain.call({ input: query }) + const queries = [] + queries.push(query) + multipleQueries.text.split('\n').map((q: string) => { + queries.push(q) + }) + console.log(JSON.stringify(queries)) + const docList: Document>[][] = [] + for (let i = 0; i < queries.length; i++) { + const resultOne = await this.baseRetriever.vectorStore.similaritySearch(queries[i], 5) + const docs: any[] = [] + resultOne.forEach((doc) => { + docs.push(doc) + }) + docList.push(docs) + } + + return this.reciprocalRankFunction(docList, 60) + } + + reciprocalRankFunction(docList: Document>[][], k: number): Document>[] { + docList.forEach((docs: Document>[]) => { + docs.forEach((doc: any, index: number) => { + let rank = index + 1 + if (doc.metadata.relevancy_score) { + doc.metadata.relevancy_score += 1 / (rank + k) + } else { + doc.metadata.relevancy_score = 1 / (rank + k) + } + }) + }) + const scoreArray: any[] = [] + docList.forEach((docs: Document>[]) => { + docs.forEach((doc: any) => { + scoreArray.push(doc.metadata.relevancy_score) + }) + }) + scoreArray.sort((a, b) => b - a) + const rerankedDocuments: Document>[] = [] + const seenScores: any[] = [] + scoreArray.forEach((score) => { + docList.forEach((docs) => { + docs.forEach((doc: any) => { + if (doc.metadata.relevancy_score === score && seenScores.indexOf(score) === -1) { + rerankedDocuments.push(doc) + seenScores.push(doc.metadata.relevancy_score) + } + }) + }) + }) + return rerankedDocuments.splice(0, this.topK) + } +} diff --git a/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg new file mode 100644 index 000000000..23c52d25e --- /dev/null +++ b/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg @@ -0,0 +1,7 @@ + + + + + + + \ No newline at end of file From d0ab21e733d03098a1faa3f0fd50b5d5baa9d381 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Fri, 29 Dec 2023 20:37:25 +0530 Subject: [PATCH 10/14] Compression Retriever: Addition of topK to Cohere Rerank Retriever --- .../CohereRerankRetriever/CohereRerank.ts | 8 +++++--- .../CohereRerankRetriever.ts | 19 ++++++++++++++++++- 2 files changed, 23 insertions(+), 4 deletions(-) diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts index 612581edb..55f3c4aad 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts @@ -5,12 +5,13 @@ import axios from 'axios' export class CohereRerank extends BaseDocumentCompressor { private cohereAPIKey: any private COHERE_API_URL = 'https://api.cohere.ai/v1/rerank' - private model: string - - constructor(cohereAPIKey: string, model: string) { + private readonly model: string + private readonly k: number + constructor(cohereAPIKey: string, model: string, k: number) { super() this.cohereAPIKey = cohereAPIKey this.model = model + this.k = k } async compressDocuments( documents: Document>[], @@ -30,6 +31,7 @@ export class CohereRerank extends BaseDocumentCompressor { } const data = { model: this.model, + topN: this.k, max_chunks_per_doc: 10, query: query, return_documents: false, diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts index 2e7090bcf..3c1872b3f 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts @@ -3,6 +3,7 @@ import { BaseRetriever } from 'langchain/schema/retriever' import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' import { getCredentialData, getCredentialParam } from '../../../src' import { CohereRerank } from './CohereRerank' +import { VectorStoreRetriever } from 'langchain/vectorstores/base' class CohereRerankRetriever_Retrievers implements INode { label: string @@ -56,6 +57,16 @@ class CohereRerankRetriever_Retrievers implements INode { ], default: 'rerank-english-v2.0', optional: true + }, + { + label: 'Top K', + name: 'topK', + description: 'Number of top results to fetch. Default to the TopK of the Base Retriever', + placeholder: '0', + type: 'number', + default: 0, + additionalParams: true, + optional: true } ] } @@ -65,8 +76,14 @@ class CohereRerankRetriever_Retrievers implements INode { const model = nodeData.inputs?.model as string const credentialData = await getCredentialData(nodeData.credential ?? '', options) const cohereApiKey = getCredentialParam('cohereApiKey', credentialData, nodeData) + const topK = nodeData.inputs?.topK as string + let k = topK ? parseFloat(topK) : 4 - const cohereCompressor = new CohereRerank(cohereApiKey, model) + if (k <= 0) { + k = (baseRetriever as VectorStoreRetriever).k + } + + const cohereCompressor = new CohereRerank(cohereApiKey, model, k) return new ContextualCompressionRetriever({ baseCompressor: cohereCompressor, baseRetriever: baseRetriever From 1bd3b5d0ee9395ed5bce48eadc5c4e58f5385099 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Sat, 30 Dec 2023 08:07:15 +0530 Subject: [PATCH 11/14] Compression Retriever: Addition of constant to RRF Retriever --- .../retrievers/RRFRetriever/RRFRetriever.ts | 16 +++++++++++++++- .../RRFRetriever/ReciprocalRankFusion.ts | 6 ++++-- 2 files changed, 19 insertions(+), 3 deletions(-) diff --git a/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts index 8d6d9d6fe..3229b3a8f 100644 --- a/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts +++ b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts @@ -57,6 +57,18 @@ class RRFRetriever_Retrievers implements INode { default: 0, additionalParams: true, optional: true + }, + { + label: 'Constant', + name: 'c', + description: + 'A constant added to the rank, controlling the balance between the importance of high-ranked items and the consideration given to lower-ranked items.\n' + + 'Default is 60', + placeholder: '60', + type: 'number', + default: 60, + additionalParams: true, + optional: true } ] } @@ -68,12 +80,14 @@ class RRFRetriever_Retrievers implements INode { const q = queryCount ? parseFloat(queryCount) : 4 const topK = nodeData.inputs?.topK as string let k = topK ? parseFloat(topK) : 4 + const constantC = nodeData.inputs?.c as string + let c = topK ? parseFloat(constantC) : 60 if (k <= 0) { k = (baseRetriever as VectorStoreRetriever).k } - const ragFusion = new ReciprocalRankFusion(llm, baseRetriever as VectorStoreRetriever, q, k) + const ragFusion = new ReciprocalRankFusion(llm, baseRetriever as VectorStoreRetriever, q, k, c) return new ContextualCompressionRetriever({ baseCompressor: ragFusion, baseRetriever: baseRetriever diff --git a/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts index 134d7c8a5..b14608fe6 100644 --- a/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts +++ b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts @@ -10,13 +10,15 @@ export class ReciprocalRankFusion extends BaseDocumentCompressor { private readonly llm: BaseLanguageModel private readonly queryCount: number private readonly topK: number + private readonly c: number private baseRetriever: VectorStoreRetriever - constructor(llm: BaseLanguageModel, baseRetriever: VectorStoreRetriever, queryCount: number, topK: number) { + constructor(llm: BaseLanguageModel, baseRetriever: VectorStoreRetriever, queryCount: number, topK: number, c: number) { super() this.queryCount = queryCount this.llm = llm this.baseRetriever = baseRetriever this.topK = topK + this.c = c } async compressDocuments( documents: Document>[], @@ -57,7 +59,7 @@ export class ReciprocalRankFusion extends BaseDocumentCompressor { docList.push(docs) } - return this.reciprocalRankFunction(docList, 60) + return this.reciprocalRankFunction(docList, this.c) } reciprocalRankFunction(docList: Document>[][], k: number): Document>[] { From fd55fa62dd6ba7d0679adb494bfb3f14070cdce3 Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Sat, 30 Dec 2023 08:08:15 +0530 Subject: [PATCH 12/14] fixes for lint failures --- .../nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts | 1 - 1 file changed, 1 deletion(-) diff --git a/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts index b14608fe6..0789ca17e 100644 --- a/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts +++ b/packages/components/nodes/retrievers/RRFRetriever/ReciprocalRankFusion.ts @@ -48,7 +48,6 @@ export class ReciprocalRankFusion extends BaseDocumentCompressor { multipleQueries.text.split('\n').map((q: string) => { queries.push(q) }) - console.log(JSON.stringify(queries)) const docList: Document>[][] = [] for (let i = 0; i < queries.length; i++) { const resultOne = await this.baseRetriever.vectorStore.similaritySearch(queries[i], 5) From 3407fa92f4e39f8d4804257224639063deced46a Mon Sep 17 00:00:00 2001 From: vinodkiran Date: Wed, 17 Jan 2024 18:29:37 +0530 Subject: [PATCH 13/14] Compression Retriever - Cohere Rerank - Add max chunks per document as optional parameter --- .../CohereRerankRetriever/CohereRerank.ts | 6 ++++-- .../CohereRerankRetriever/CohereRerankRetriever.ts | 13 ++++++++++++- 2 files changed, 16 insertions(+), 3 deletions(-) diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts index 55f3c4aad..f74b83655 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts @@ -7,11 +7,13 @@ export class CohereRerank extends BaseDocumentCompressor { private COHERE_API_URL = 'https://api.cohere.ai/v1/rerank' private readonly model: string private readonly k: number - constructor(cohereAPIKey: string, model: string, k: number) { + private readonly maxChunksPerDoc: number + constructor(cohereAPIKey: string, model: string, k: number, maxChunksPerDoc: number) { super() this.cohereAPIKey = cohereAPIKey this.model = model this.k = k + this.maxChunksPerDoc = maxChunksPerDoc } async compressDocuments( documents: Document>[], @@ -32,7 +34,7 @@ export class CohereRerank extends BaseDocumentCompressor { const data = { model: this.model, topN: this.k, - max_chunks_per_doc: 10, + max_chunks_per_doc: this.maxChunksPerDoc, query: query, return_documents: false, documents: documents.map((doc) => doc.pageContent) diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts index 3c1872b3f..ca89ca771 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts @@ -67,6 +67,15 @@ class CohereRerankRetriever_Retrievers implements INode { default: 0, additionalParams: true, optional: true + }, + { + label: 'Max Chunks Per Document', + name: 'maxChunksPerDoc', + placeholder: '10', + type: 'number', + default: 10, + additionalParams: true, + optional: true } ] } @@ -78,12 +87,14 @@ class CohereRerankRetriever_Retrievers implements INode { const cohereApiKey = getCredentialParam('cohereApiKey', credentialData, nodeData) const topK = nodeData.inputs?.topK as string let k = topK ? parseFloat(topK) : 4 + const maxChunks = nodeData.inputs?.maxChunksPerDoc as string + let max = maxChunks ? parseInt(maxChunks) : 10 if (k <= 0) { k = (baseRetriever as VectorStoreRetriever).k } - const cohereCompressor = new CohereRerank(cohereApiKey, model, k) + const cohereCompressor = new CohereRerank(cohereApiKey, model, k, max) return new ContextualCompressionRetriever({ baseCompressor: cohereCompressor, baseRetriever: baseRetriever From 1bf7944776c83ceb3903fe9d33dba82f2acb1fdf Mon Sep 17 00:00:00 2001 From: Henry Date: Wed, 17 Jan 2024 15:55:56 +0000 Subject: [PATCH 14/14] update retrievers and add mmr to other vector stores --- .../CohereRerankRetriever/Cohere.svg | 1 + .../CohereRerankRetriever/CohereRerank.ts | 7 +- .../CohereRerankRetriever.ts | 74 ++++++++++++--- .../compressionRetriever.svg | 7 -- .../EmbeddingsFilterRetriever.ts | 62 ++++++++++--- .../retrievers/HydeRetriever/HydeRetriever.ts | 50 ++++++++++- .../LLMFilterCompressionRetriever.ts | 62 ++++++++++--- .../compressionRetriever.svg | 7 -- .../LLMFilterRetriever/llmFilterRetriever.svg | 1 + .../retrievers/RRFRetriever/RRFRetriever.ts | 64 ++++++++++--- .../RRFRetriever/compressionRetriever.svg | 7 -- .../retrievers/RRFRetriever/rrfRetriever.svg | 1 + .../SimilarityThresholdRetriever.ts | 21 +++-- .../nodes/vectorstores/Astra/Astra.ts | 16 +--- .../vectorstores/MongoDBAtlas/MongoDBAtlas.ts | 15 +--- .../nodes/vectorstores/Pinecone/Pinecone.ts | 2 +- .../marketplaces/chatflows/AutoGPT.json | 46 +++++++++- .../marketplaces/chatflows/BabyAGI.json | 46 +++++++++- .../Conversational Retrieval Agent.json | 46 +++++++++- .../Conversational Retrieval QA Chain.json | 46 +++++++++- .../chatflows/Metadata Filter.json | 46 +++++++++- .../chatflows/Multi Retrieval QA Chain.json | 90 ++++++++++++++++++- .../marketplaces/chatflows/WebPage QnA.json | 46 +++++++++- 23 files changed, 642 insertions(+), 121 deletions(-) create mode 100644 packages/components/nodes/retrievers/CohereRerankRetriever/Cohere.svg delete mode 100644 packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg delete mode 100644 packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg create mode 100644 packages/components/nodes/retrievers/LLMFilterRetriever/llmFilterRetriever.svg delete mode 100644 packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg create mode 100644 packages/components/nodes/retrievers/RRFRetriever/rrfRetriever.svg diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/Cohere.svg b/packages/components/nodes/retrievers/CohereRerankRetriever/Cohere.svg new file mode 100644 index 000000000..88bcabe34 --- /dev/null +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/Cohere.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts index 55f3c4aad..e70c044f6 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerank.ts @@ -7,11 +7,14 @@ export class CohereRerank extends BaseDocumentCompressor { private COHERE_API_URL = 'https://api.cohere.ai/v1/rerank' private readonly model: string private readonly k: number - constructor(cohereAPIKey: string, model: string, k: number) { + private readonly max_chunks_per_doc: number + + constructor(cohereAPIKey: string, model: string, k: number, max_chunks_per_doc: number) { super() this.cohereAPIKey = cohereAPIKey this.model = model this.k = k + this.max_chunks_per_doc = max_chunks_per_doc } async compressDocuments( documents: Document>[], @@ -32,8 +35,8 @@ export class CohereRerank extends BaseDocumentCompressor { const data = { model: this.model, topN: this.k, - max_chunks_per_doc: 10, query: query, + max_chunks_per_doc: this.max_chunks_per_doc, return_documents: false, documents: documents.map((doc) => doc.pageContent) } diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts index 3c1872b3f..442fdc7a6 100644 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts +++ b/packages/components/nodes/retrievers/CohereRerankRetriever/CohereRerankRetriever.ts @@ -1,7 +1,7 @@ import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { BaseRetriever } from 'langchain/schema/retriever' import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' -import { getCredentialData, getCredentialParam } from '../../../src' +import { getCredentialData, getCredentialParam, handleEscapeCharacters } from '../../../src' import { CohereRerank } from './CohereRerank' import { VectorStoreRetriever } from 'langchain/vectorstores/base' @@ -15,16 +15,16 @@ class CohereRerankRetriever_Retrievers implements INode { category: string baseClasses: string[] inputs: INodeParams[] - outputs: INodeOutputsValue[] credential: INodeParams badge: string + outputs: INodeOutputsValue[] constructor() { this.label = 'Cohere Rerank Retriever' this.name = 'cohereRerankRetriever' this.version = 1.0 this.type = 'Cohere Rerank Retriever' - this.icon = 'compressionRetriever.svg' + this.icon = 'Cohere.svg' this.category = 'Retrievers' this.badge = 'NEW' this.description = 'Cohere Rerank indexes the documents from most to least semantically relevant to the query.' @@ -37,7 +37,7 @@ class CohereRerankRetriever_Retrievers implements INode { } this.inputs = [ { - label: 'Base Retriever', + label: 'Vector Store Retriever', name: 'baseRetriever', type: 'VectorStoreRetriever' }, @@ -58,36 +58,84 @@ class CohereRerankRetriever_Retrievers implements INode { default: 'rerank-english-v2.0', optional: true }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true + }, { label: 'Top K', name: 'topK', description: 'Number of top results to fetch. Default to the TopK of the Base Retriever', - placeholder: '0', + placeholder: '4', + type: 'number', + additionalParams: true, + optional: true + }, + { + label: 'Max Chunks Per Doc', + name: 'maxChunksPerDoc', + description: 'The maximum number of chunks to produce internally from a document. Default to 10', + placeholder: '10', type: 'number', - default: 0, additionalParams: true, optional: true } ] + this.outputs = [ + { + label: 'Cohere Rerank Retriever', + name: 'retriever', + baseClasses: this.baseClasses + }, + { + label: 'Document', + name: 'document', + baseClasses: ['Document'] + }, + { + label: 'Text', + name: 'text', + baseClasses: ['string', 'json'] + } + ] } - async init(nodeData: INodeData, _: string, options: ICommonObject): Promise { + async init(nodeData: INodeData, input: string, options: ICommonObject): Promise { const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever const model = nodeData.inputs?.model as string + const query = nodeData.inputs?.query as string const credentialData = await getCredentialData(nodeData.credential ?? '', options) const cohereApiKey = getCredentialParam('cohereApiKey', credentialData, nodeData) const topK = nodeData.inputs?.topK as string - let k = topK ? parseFloat(topK) : 4 + const k = topK ? parseFloat(topK) : (baseRetriever as VectorStoreRetriever).k ?? 4 + const maxChunksPerDoc = nodeData.inputs?.maxChunksPerDoc as string + const max_chunks_per_doc = maxChunksPerDoc ? parseFloat(maxChunksPerDoc) : 10 + const output = nodeData.outputs?.output as string - if (k <= 0) { - k = (baseRetriever as VectorStoreRetriever).k - } + const cohereCompressor = new CohereRerank(cohereApiKey, model, k, max_chunks_per_doc) - const cohereCompressor = new CohereRerank(cohereApiKey, model, k) - return new ContextualCompressionRetriever({ + const retriever = new ContextualCompressionRetriever({ baseCompressor: cohereCompressor, baseRetriever: baseRetriever }) + + if (output === 'retriever') return retriever + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) + else if (output === 'text') { + let finaltext = '' + + const docs = await retriever.getRelevantDocuments(query ? query : input) + + for (const doc of docs) finaltext += `${doc.pageContent}\n` + + return handleEscapeCharacters(finaltext, false) + } + + return retriever } } diff --git a/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg deleted file mode 100644 index 23c52d25e..000000000 --- a/packages/components/nodes/retrievers/CohereRerankRetriever/compressionRetriever.svg +++ /dev/null @@ -1,7 +0,0 @@ - - - - - - - \ No newline at end of file diff --git a/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts index d373704c0..d1049fa49 100644 --- a/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts +++ b/packages/components/nodes/retrievers/EmbeddingsFilterRetriever/EmbeddingsFilterRetriever.ts @@ -3,6 +3,7 @@ import { BaseRetriever } from 'langchain/schema/retriever' import { Embeddings } from 'langchain/embeddings/base' import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' import { EmbeddingsFilter } from 'langchain/retrievers/document_compressors/embeddings_filter' +import { handleEscapeCharacters } from '../../../src/utils' class EmbeddingsFilterRetriever_Retrievers implements INode { label: string @@ -29,15 +30,22 @@ class EmbeddingsFilterRetriever_Retrievers implements INode { this.baseClasses = [this.type, 'BaseRetriever'] this.inputs = [ { - label: 'Base Retriever', + label: 'Vector Store Retriever', name: 'baseRetriever', type: 'VectorStoreRetriever' }, { label: 'Embeddings', name: 'embeddings', - type: 'Embeddings', - optional: false + type: 'Embeddings' + }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true }, { label: 'Similarity Threshold', @@ -61,36 +69,64 @@ class EmbeddingsFilterRetriever_Retrievers implements INode { additionalParams: true } ] + this.outputs = [ + { + label: 'Embeddings Filter Retriever', + name: 'retriever', + baseClasses: this.baseClasses + }, + { + label: 'Document', + name: 'document', + baseClasses: ['Document'] + }, + { + label: 'Text', + name: 'text', + baseClasses: ['string', 'json'] + } + ] } - async init(nodeData: INodeData): Promise { + async init(nodeData: INodeData, input: string): Promise { const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever const embeddings = nodeData.inputs?.embeddings as Embeddings + const query = nodeData.inputs?.query as string const similarityThreshold = nodeData.inputs?.similarityThreshold as string const k = nodeData.inputs?.k as string + const output = nodeData.outputs?.output as string if (k === undefined && similarityThreshold === undefined) { throw new Error(`Must specify one of "k" or "similarity_threshold".`) } - let similarityThresholdNumber = 0.8 - if (similarityThreshold) { - similarityThresholdNumber = parseFloat(similarityThreshold) - } - let kNumber = 0.8 - if (k) { - kNumber = parseFloat(k) - } + const similarityThresholdNumber = similarityThreshold ? parseFloat(similarityThreshold) : 0.8 + const kNumber = k ? parseFloat(k) : undefined + const baseCompressor = new EmbeddingsFilter({ embeddings: embeddings, similarityThreshold: similarityThresholdNumber, k: kNumber }) - return new ContextualCompressionRetriever({ + const retriever = new ContextualCompressionRetriever({ baseCompressor, baseRetriever: baseRetriever }) + + if (output === 'retriever') return retriever + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) + else if (output === 'text') { + let finaltext = '' + + const docs = await retriever.getRelevantDocuments(query ? query : input) + + for (const doc of docs) finaltext += `${doc.pageContent}\n` + + return handleEscapeCharacters(finaltext, false) + } + + return retriever } } diff --git a/packages/components/nodes/retrievers/HydeRetriever/HydeRetriever.ts b/packages/components/nodes/retrievers/HydeRetriever/HydeRetriever.ts index 10d9a6e7a..10fff7646 100644 --- a/packages/components/nodes/retrievers/HydeRetriever/HydeRetriever.ts +++ b/packages/components/nodes/retrievers/HydeRetriever/HydeRetriever.ts @@ -1,8 +1,9 @@ import { VectorStore } from 'langchain/vectorstores/base' -import { INode, INodeData, INodeParams } from '../../../src/Interface' +import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { HydeRetriever, HydeRetrieverOptions, PromptKey } from 'langchain/retrievers/hyde' import { BaseLanguageModel } from 'langchain/base_language' import { PromptTemplate } from 'langchain/prompts' +import { handleEscapeCharacters } from '../../../src/utils' class HydeRetriever_Retrievers implements INode { label: string @@ -14,11 +15,12 @@ class HydeRetriever_Retrievers implements INode { category: string baseClasses: string[] inputs: INodeParams[] + outputs: INodeOutputsValue[] constructor() { - this.label = 'Hyde Retriever' + this.label = 'HyDE Retriever' this.name = 'HydeRetriever' - this.version = 2.0 + this.version = 3.0 this.type = 'HydeRetriever' this.icon = 'hyderetriever.svg' this.category = 'Retrievers' @@ -35,6 +37,14 @@ class HydeRetriever_Retrievers implements INode { name: 'vectorStore', type: 'VectorStore' }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true + }, { label: 'Select Defined Prompt', name: 'promptKey', @@ -121,15 +131,34 @@ Passage:` optional: true } ] + this.outputs = [ + { + label: 'HyDE Retriever', + name: 'retriever', + baseClasses: this.baseClasses + }, + { + label: 'Document', + name: 'document', + baseClasses: ['Document'] + }, + { + label: 'Text', + name: 'text', + baseClasses: ['string', 'json'] + } + ] } - async init(nodeData: INodeData): Promise { + async init(nodeData: INodeData, input: string): Promise { const llm = nodeData.inputs?.model as BaseLanguageModel const vectorStore = nodeData.inputs?.vectorStore as VectorStore const promptKey = nodeData.inputs?.promptKey as PromptKey const customPrompt = nodeData.inputs?.customPrompt as string + const query = nodeData.inputs?.query as string const topK = nodeData.inputs?.topK as string const k = topK ? parseFloat(topK) : 4 + const output = nodeData.outputs?.output as string const obj: HydeRetrieverOptions = { llm, @@ -141,6 +170,19 @@ Passage:` else if (promptKey) obj.promptTemplate = promptKey const retriever = new HydeRetriever(obj) + + if (output === 'retriever') return retriever + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) + else if (output === 'text') { + let finaltext = '' + + const docs = await retriever.getRelevantDocuments(query ? query : input) + + for (const doc of docs) finaltext += `${doc.pageContent}\n` + + return handleEscapeCharacters(finaltext, false) + } + return retriever } } diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts index e044468f4..6b710cf30 100644 --- a/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts +++ b/packages/components/nodes/retrievers/LLMFilterRetriever/LLMFilterCompressionRetriever.ts @@ -3,6 +3,7 @@ import { BaseRetriever } from 'langchain/schema/retriever' import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' import { BaseLanguageModel } from 'langchain/base_language' import { LLMChainExtractor } from 'langchain/retrievers/document_compressors/chain_extract' +import { handleEscapeCharacters } from '../../../src/utils' class LLMFilterCompressionRetriever_Retrievers implements INode { label: string @@ -22,7 +23,7 @@ class LLMFilterCompressionRetriever_Retrievers implements INode { this.name = 'llmFilterRetriever' this.version = 1.0 this.type = 'LLMFilterRetriever' - this.icon = 'compressionRetriever.svg' + this.icon = 'llmFilterRetriever.svg' this.category = 'Retrievers' this.badge = 'NEW' this.description = @@ -30,30 +31,69 @@ class LLMFilterCompressionRetriever_Retrievers implements INode { this.baseClasses = [this.type, 'BaseRetriever'] this.inputs = [ { - label: 'Base Retriever', + label: 'Vector Store Retriever', name: 'baseRetriever', type: 'VectorStoreRetriever' }, { label: 'Language Model', name: 'model', - type: 'BaseLanguageModel', - optional: true + type: 'BaseLanguageModel' + }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true + } + ] + this.outputs = [ + { + label: 'LLM Filter Retriever', + name: 'retriever', + baseClasses: this.baseClasses + }, + { + label: 'Document', + name: 'document', + baseClasses: ['Document'] + }, + { + label: 'Text', + name: 'text', + baseClasses: ['string', 'json'] } ] } - async init(nodeData: INodeData): Promise { + async init(nodeData: INodeData, input: string): Promise { const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever const model = nodeData.inputs?.model as BaseLanguageModel + const query = nodeData.inputs?.query as string + const output = nodeData.outputs?.output as string - if (model) { - return new ContextualCompressionRetriever({ - baseCompressor: LLMChainExtractor.fromLLM(model), - baseRetriever: baseRetriever - }) + if (!model) throw new Error('There must be a LLM model connected to LLM Filter Retriever') + + const retriever = new ContextualCompressionRetriever({ + baseCompressor: LLMChainExtractor.fromLLM(model), + baseRetriever: baseRetriever + }) + + if (output === 'retriever') return retriever + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) + else if (output === 'text') { + let finaltext = '' + + const docs = await retriever.getRelevantDocuments(query ? query : input) + + for (const doc of docs) finaltext += `${doc.pageContent}\n` + + return handleEscapeCharacters(finaltext, false) } - return {} + + return retriever } } diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg deleted file mode 100644 index 23c52d25e..000000000 --- a/packages/components/nodes/retrievers/LLMFilterRetriever/compressionRetriever.svg +++ /dev/null @@ -1,7 +0,0 @@ - - - - - - - \ No newline at end of file diff --git a/packages/components/nodes/retrievers/LLMFilterRetriever/llmFilterRetriever.svg b/packages/components/nodes/retrievers/LLMFilterRetriever/llmFilterRetriever.svg new file mode 100644 index 000000000..d3f4d15f4 --- /dev/null +++ b/packages/components/nodes/retrievers/LLMFilterRetriever/llmFilterRetriever.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts index 3229b3a8f..ed15ed243 100644 --- a/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts +++ b/packages/components/nodes/retrievers/RRFRetriever/RRFRetriever.ts @@ -1,9 +1,10 @@ -import { INode, INodeData, INodeParams } from '../../../src/Interface' +import { INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { BaseLanguageModel } from 'langchain/base_language' import { ContextualCompressionRetriever } from 'langchain/retrievers/contextual_compression' import { BaseRetriever } from 'langchain/schema/retriever' import { ReciprocalRankFusion } from './ReciprocalRankFusion' import { VectorStoreRetriever } from 'langchain/vectorstores/base' +import { handleEscapeCharacters } from '../../../src/utils' class RRFRetriever_Retrievers implements INode { label: string @@ -16,20 +17,21 @@ class RRFRetriever_Retrievers implements INode { baseClasses: string[] inputs: INodeParams[] badge: string + outputs: INodeOutputsValue[] constructor() { this.label = 'Reciprocal Rank Fusion Retriever' this.name = 'RRFRetriever' - this.version = 2.0 + this.version = 1.0 this.type = 'RRFRetriever' this.badge = 'NEW' - this.icon = 'compressionRetriever.svg' + this.icon = 'rrfRetriever.svg' this.category = 'Retrievers' this.description = 'Reciprocal Rank Fusion to re-rank search results by multiple query generation.' this.baseClasses = [this.type, 'BaseRetriever'] this.inputs = [ { - label: 'Base Retriever', + label: 'Vector Store Retriever', name: 'baseRetriever', type: 'VectorStoreRetriever' }, @@ -38,6 +40,14 @@ class RRFRetriever_Retrievers implements INode { name: 'model', type: 'BaseLanguageModel' }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true + }, { label: 'Query Count', name: 'queryCount', @@ -54,7 +64,6 @@ class RRFRetriever_Retrievers implements INode { description: 'Number of top results to fetch. Default to the TopK of the Base Retriever', placeholder: '0', type: 'number', - default: 0, additionalParams: true, optional: true }, @@ -71,27 +80,56 @@ class RRFRetriever_Retrievers implements INode { optional: true } ] + this.outputs = [ + { + label: 'Reciprocal Rank Fusion Retriever', + name: 'retriever', + baseClasses: this.baseClasses + }, + { + label: 'Document', + name: 'document', + baseClasses: ['Document'] + }, + { + label: 'Text', + name: 'text', + baseClasses: ['string', 'json'] + } + ] } - async init(nodeData: INodeData): Promise { + async init(nodeData: INodeData, input: string): Promise { const llm = nodeData.inputs?.model as BaseLanguageModel const baseRetriever = nodeData.inputs?.baseRetriever as BaseRetriever + const query = nodeData.inputs?.query as string const queryCount = nodeData.inputs?.queryCount as string const q = queryCount ? parseFloat(queryCount) : 4 const topK = nodeData.inputs?.topK as string - let k = topK ? parseFloat(topK) : 4 + const k = topK ? parseFloat(topK) : (baseRetriever as VectorStoreRetriever).k ?? 4 const constantC = nodeData.inputs?.c as string - let c = topK ? parseFloat(constantC) : 60 - - if (k <= 0) { - k = (baseRetriever as VectorStoreRetriever).k - } + const c = topK ? parseFloat(constantC) : 60 + const output = nodeData.outputs?.output as string const ragFusion = new ReciprocalRankFusion(llm, baseRetriever as VectorStoreRetriever, q, k, c) - return new ContextualCompressionRetriever({ + const retriever = new ContextualCompressionRetriever({ baseCompressor: ragFusion, baseRetriever: baseRetriever }) + + if (output === 'retriever') return retriever + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) + else if (output === 'text') { + let finaltext = '' + + const docs = await retriever.getRelevantDocuments(query ? query : input) + + for (const doc of docs) finaltext += `${doc.pageContent}\n` + + return handleEscapeCharacters(finaltext, false) + } + + return retriever } } diff --git a/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg b/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg deleted file mode 100644 index 23c52d25e..000000000 --- a/packages/components/nodes/retrievers/RRFRetriever/compressionRetriever.svg +++ /dev/null @@ -1,7 +0,0 @@ - - - - - - - \ No newline at end of file diff --git a/packages/components/nodes/retrievers/RRFRetriever/rrfRetriever.svg b/packages/components/nodes/retrievers/RRFRetriever/rrfRetriever.svg new file mode 100644 index 000000000..56fbcc5a1 --- /dev/null +++ b/packages/components/nodes/retrievers/RRFRetriever/rrfRetriever.svg @@ -0,0 +1 @@ + \ No newline at end of file diff --git a/packages/components/nodes/retrievers/SimilarityThresholdRetriever/SimilarityThresholdRetriever.ts b/packages/components/nodes/retrievers/SimilarityThresholdRetriever/SimilarityThresholdRetriever.ts index a9f4b3d87..5f5a9ed0d 100644 --- a/packages/components/nodes/retrievers/SimilarityThresholdRetriever/SimilarityThresholdRetriever.ts +++ b/packages/components/nodes/retrievers/SimilarityThresholdRetriever/SimilarityThresholdRetriever.ts @@ -18,7 +18,7 @@ class SimilarityThresholdRetriever_Retrievers implements INode { constructor() { this.label = 'Similarity Score Threshold Retriever' this.name = 'similarityThresholdRetriever' - this.version = 1.0 + this.version = 2.0 this.type = 'SimilarityThresholdRetriever' this.icon = 'similaritythreshold.svg' this.category = 'Retrievers' @@ -30,6 +30,14 @@ class SimilarityThresholdRetriever_Retrievers implements INode { name: 'vectorStore', type: 'VectorStore' }, + { + label: 'Query', + name: 'query', + type: 'string', + description: 'Query to retrieve documents from retriever. If not specified, user question will be used', + optional: true, + acceptVariable: true + }, { label: 'Minimum Similarity Score (%)', name: 'minSimilarityScore', @@ -44,7 +52,8 @@ class SimilarityThresholdRetriever_Retrievers implements INode { description: `The maximum number of results to fetch`, type: 'number', default: 20, - step: 1 + step: 1, + additionalParams: true }, { label: 'K Increment', @@ -52,7 +61,8 @@ class SimilarityThresholdRetriever_Retrievers implements INode { description: `How much to increase K by each time. It'll fetch N results, then N + kIncrement, then N + kIncrement * 2, etc.`, type: 'number', default: 2, - step: 1 + step: 1, + additionalParams: true } ] this.outputs = [ @@ -77,6 +87,7 @@ class SimilarityThresholdRetriever_Retrievers implements INode { async init(nodeData: INodeData, input: string): Promise { const vectorStore = nodeData.inputs?.vectorStore as VectorStore const minSimilarityScore = nodeData.inputs?.minSimilarityScore as number + const query = nodeData.inputs?.query as string const maxK = nodeData.inputs?.maxK as string const kIncrement = nodeData.inputs?.kIncrement as string @@ -89,11 +100,11 @@ class SimilarityThresholdRetriever_Retrievers implements INode { }) if (output === 'retriever') return retriever - else if (output === 'document') return await retriever.getRelevantDocuments(input) + else if (output === 'document') return await retriever.getRelevantDocuments(query ? query : input) else if (output === 'text') { let finaltext = '' - const docs = await retriever.getRelevantDocuments(input) + const docs = await retriever.getRelevantDocuments(query ? query : input) for (const doc of docs) finaltext += `${doc.pageContent}\n` diff --git a/packages/components/nodes/vectorstores/Astra/Astra.ts b/packages/components/nodes/vectorstores/Astra/Astra.ts index 865f10446..edaadc9c0 100644 --- a/packages/components/nodes/vectorstores/Astra/Astra.ts +++ b/packages/components/nodes/vectorstores/Astra/Astra.ts @@ -4,6 +4,7 @@ import { Document } from 'langchain/document' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData } from '../../../src/utils' import { AstraDBVectorStore, AstraLibArgs } from '@langchain/community/vectorstores/astradb' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class Astra_VectorStores implements INode { label: string @@ -26,7 +27,7 @@ class Astra_VectorStores implements INode { this.type = 'Astra' this.icon = 'astra.svg' this.category = 'Vector Stores' - this.description = `Upsert embedded data and perform similarity search upon query using DataStax Astra DB, a serverless vector database that’s perfect for managing mission-critical AI workloads` + this.description = `Upsert embedded data and perform similarity or mmr search upon query using DataStax Astra DB, a serverless vector database that’s perfect for managing mission-critical AI workloads` this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever'] this.badge = 'NEW' this.credential = { @@ -74,6 +75,7 @@ class Astra_VectorStores implements INode { optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'Astra Retriever', @@ -139,9 +141,6 @@ class Astra_VectorStores implements INode { const embeddings = nodeData.inputs?.embeddings as Embeddings const vectorDimension = nodeData.inputs?.vectorDimension as number const similarityMetric = nodeData.inputs?.similarityMetric as 'cosine' | 'euclidean' | 'dot_product' | undefined - const output = nodeData.outputs?.output as string - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 const credentialData = await getCredentialData(nodeData.credential ?? '', options) @@ -176,14 +175,7 @@ class Astra_VectorStores implements INode { const vectorStore = await AstraDBVectorStore.fromExistingIndex(embeddings, astraConfig) - if (output === 'retriever') { - const retriever = vectorStore.asRetriever(k) - return retriever - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } diff --git a/packages/components/nodes/vectorstores/MongoDBAtlas/MongoDBAtlas.ts b/packages/components/nodes/vectorstores/MongoDBAtlas/MongoDBAtlas.ts index 9bc23f104..6ba7199f0 100644 --- a/packages/components/nodes/vectorstores/MongoDBAtlas/MongoDBAtlas.ts +++ b/packages/components/nodes/vectorstores/MongoDBAtlas/MongoDBAtlas.ts @@ -5,6 +5,7 @@ import { Embeddings } from 'langchain/embeddings/base' import { Document } from 'langchain/document' import { ICommonObject, INode, INodeData, INodeOutputsValue, INodeParams } from '../../../src/Interface' import { getBaseClasses, getCredentialData, getCredentialParam } from '../../../src/utils' +import { addMMRInputParams, resolveVectorStoreOrRetriever } from '../VectorStoreUtils' class MongoDBAtlas_VectorStores implements INode { label: string @@ -24,7 +25,7 @@ class MongoDBAtlas_VectorStores implements INode { this.label = 'MongoDB Atlas' this.name = 'mongoDBAtlas' this.version = 1.0 - this.description = `Upsert embedded data and perform similarity search upon query using MongoDB Atlas, a managed cloud mongodb database` + this.description = `Upsert embedded data and perform similarity or mmr search upon query using MongoDB Atlas, a managed cloud mongodb database` this.type = 'MongoDB Atlas' this.icon = 'mongodb.svg' this.category = 'Vector Stores' @@ -95,6 +96,7 @@ class MongoDBAtlas_VectorStores implements INode { optional: true } ] + addMMRInputParams(this.inputs) this.outputs = [ { label: 'MongoDB Retriever', @@ -162,9 +164,6 @@ class MongoDBAtlas_VectorStores implements INode { let textKey = nodeData.inputs?.textKey as string let embeddingKey = nodeData.inputs?.embeddingKey as string const embeddings = nodeData.inputs?.embeddings as Embeddings - const topK = nodeData.inputs?.topK as string - const k = topK ? parseFloat(topK) : 4 - const output = nodeData.outputs?.output as string let mongoDBConnectUrl = getCredentialParam('mongoDBConnectUrl', credentialData, nodeData) @@ -181,13 +180,7 @@ class MongoDBAtlas_VectorStores implements INode { embeddingKey }) - if (output === 'retriever') { - return vectorStore.asRetriever(k) - } else if (output === 'vectorStore') { - ;(vectorStore as any).k = k - return vectorStore - } - return vectorStore + return resolveVectorStoreOrRetriever(nodeData, vectorStore) } } diff --git a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts index 4b91a9b54..6623b1a26 100644 --- a/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts +++ b/packages/components/nodes/vectorstores/Pinecone/Pinecone.ts @@ -24,7 +24,7 @@ class Pinecone_VectorStores implements INode { constructor() { this.label = 'Pinecone' this.name = 'pinecone' - this.version = 3.0 + this.version = 2.0 this.type = 'Pinecone' this.icon = 'pinecone.svg' this.category = 'Vector Stores' diff --git a/packages/server/marketplaces/chatflows/AutoGPT.json b/packages/server/marketplaces/chatflows/AutoGPT.json index 150fe17eb..0062cd43f 100644 --- a/packages/server/marketplaces/chatflows/AutoGPT.json +++ b/packages/server/marketplaces/chatflows/AutoGPT.json @@ -511,7 +511,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -552,6 +552,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -576,7 +615,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/BabyAGI.json b/packages/server/marketplaces/chatflows/BabyAGI.json index ab387205e..81e3f2307 100644 --- a/packages/server/marketplaces/chatflows/BabyAGI.json +++ b/packages/server/marketplaces/chatflows/BabyAGI.json @@ -166,7 +166,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -207,6 +207,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -231,7 +270,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/Conversational Retrieval Agent.json b/packages/server/marketplaces/chatflows/Conversational Retrieval Agent.json index 0e9e41bdd..4378a47d6 100644 --- a/packages/server/marketplaces/chatflows/Conversational Retrieval Agent.json +++ b/packages/server/marketplaces/chatflows/Conversational Retrieval Agent.json @@ -301,7 +301,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -342,6 +342,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -366,7 +405,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/Conversational Retrieval QA Chain.json b/packages/server/marketplaces/chatflows/Conversational Retrieval QA Chain.json index e2fd64210..253a1dfc7 100644 --- a/packages/server/marketplaces/chatflows/Conversational Retrieval QA Chain.json +++ b/packages/server/marketplaces/chatflows/Conversational Retrieval QA Chain.json @@ -541,7 +541,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -582,6 +582,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -606,7 +645,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/Metadata Filter.json b/packages/server/marketplaces/chatflows/Metadata Filter.json index abd85d366..f7b2fbfb1 100644 --- a/packages/server/marketplaces/chatflows/Metadata Filter.json +++ b/packages/server/marketplaces/chatflows/Metadata Filter.json @@ -625,7 +625,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -666,6 +666,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -690,7 +729,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "{\"id\":{\"$in\":[\"doc1\",\"doc2\"]}}", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/Multi Retrieval QA Chain.json b/packages/server/marketplaces/chatflows/Multi Retrieval QA Chain.json index 5388d9657..e86b28c93 100644 --- a/packages/server/marketplaces/chatflows/Multi Retrieval QA Chain.json +++ b/packages/server/marketplaces/chatflows/Multi Retrieval QA Chain.json @@ -560,7 +560,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -601,6 +601,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -625,7 +664,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { @@ -840,6 +882,45 @@ "additionalParams": true, "optional": true, "id": "supabase_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -865,7 +946,10 @@ "tableName": "", "queryName": "", "supabaseMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ { diff --git a/packages/server/marketplaces/chatflows/WebPage QnA.json b/packages/server/marketplaces/chatflows/WebPage QnA.json index 1b1d8de66..df05feef4 100644 --- a/packages/server/marketplaces/chatflows/WebPage QnA.json +++ b/packages/server/marketplaces/chatflows/WebPage QnA.json @@ -643,7 +643,7 @@ "type": "Pinecone", "baseClasses": ["Pinecone", "VectorStoreRetriever", "BaseRetriever"], "category": "Vector Stores", - "description": "Upsert embedded data and perform similarity search upon query using Pinecone, a leading fully managed hosted vector database", + "description": "Upsert embedded data and perform similarity or mmr search using Pinecone, a leading fully managed hosted vector database", "inputParams": [ { "label": "Connect Credential", @@ -684,6 +684,45 @@ "additionalParams": true, "optional": true, "id": "pinecone_0-input-topK-number" + }, + { + "label": "Search Type", + "name": "searchType", + "type": "options", + "default": "similarity", + "options": [ + { + "label": "Similarity", + "name": "similarity" + }, + { + "label": "Max Marginal Relevance", + "name": "mmr" + } + ], + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-searchType-options" + }, + { + "label": "Fetch K (for MMR Search)", + "name": "fetchK", + "description": "Number of initial documents to fetch for MMR reranking. Default to 20. Used only when the search type is MMR", + "placeholder": "20", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-fetchK-number" + }, + { + "label": "Lambda (for MMR Search)", + "name": "lambda", + "description": "Number between 0 and 1 that determines the degree of diversity among the results, where 0 corresponds to maximum diversity and 1 to minimum diversity. Used only when the search type is MMR", + "placeholder": "0.5", + "type": "number", + "additionalParams": true, + "optional": true, + "id": "pinecone_0-input-lambda-number" } ], "inputAnchors": [ @@ -708,7 +747,10 @@ "pineconeIndex": "", "pineconeNamespace": "", "pineconeMetadataFilter": "", - "topK": "" + "topK": "", + "searchType": "similarity", + "fetchK": "", + "lambda": "" }, "outputAnchors": [ {