147 lines
5.1 KiB
TypeScript
147 lines
5.1 KiB
TypeScript
import { flatten } from 'lodash'
|
|
import { Document } from '@langchain/core/documents'
|
|
import { FaissStore } from '@langchain/community/vectorstores/faiss'
|
|
import { Embeddings } from '@langchain/core/embeddings'
|
|
import { INode, INodeData, INodeOutputsValue, INodeParams, IndexingResult } from '../../../src/Interface'
|
|
import { getBaseClasses } from '../../../src/utils'
|
|
|
|
class Faiss_VectorStores implements INode {
|
|
label: string
|
|
name: string
|
|
version: number
|
|
description: string
|
|
type: string
|
|
icon: string
|
|
category: string
|
|
badge: string
|
|
baseClasses: string[]
|
|
inputs: INodeParams[]
|
|
outputs: INodeOutputsValue[]
|
|
|
|
constructor() {
|
|
this.label = 'Faiss'
|
|
this.name = 'faiss'
|
|
this.version = 1.0
|
|
this.type = 'Faiss'
|
|
this.icon = 'faiss.svg'
|
|
this.category = 'Vector Stores'
|
|
this.description = 'Upsert embedded data and perform similarity search upon query using Faiss library from Meta'
|
|
this.baseClasses = [this.type, 'VectorStoreRetriever', 'BaseRetriever']
|
|
this.inputs = [
|
|
{
|
|
label: 'Document',
|
|
name: 'document',
|
|
type: 'Document',
|
|
list: true,
|
|
optional: true
|
|
},
|
|
{
|
|
label: 'Embeddings',
|
|
name: 'embeddings',
|
|
type: 'Embeddings'
|
|
},
|
|
{
|
|
label: 'Base Path to load',
|
|
name: 'basePath',
|
|
description: 'Path to load faiss.index file',
|
|
placeholder: `C:\\Users\\User\\Desktop`,
|
|
type: 'string'
|
|
},
|
|
{
|
|
label: 'Top K',
|
|
name: 'topK',
|
|
description: 'Number of top results to fetch. Default to 4',
|
|
placeholder: '4',
|
|
type: 'number',
|
|
additionalParams: true,
|
|
optional: true
|
|
}
|
|
]
|
|
this.outputs = [
|
|
{
|
|
label: 'Faiss Retriever',
|
|
name: 'retriever',
|
|
baseClasses: this.baseClasses
|
|
},
|
|
{
|
|
label: 'Faiss Vector Store',
|
|
name: 'vectorStore',
|
|
baseClasses: [this.type, ...getBaseClasses(FaissStore)]
|
|
}
|
|
]
|
|
}
|
|
|
|
//@ts-ignore
|
|
vectorStoreMethods = {
|
|
async upsert(nodeData: INodeData): Promise<Partial<IndexingResult>> {
|
|
const docs = nodeData.inputs?.document as Document[]
|
|
const embeddings = nodeData.inputs?.embeddings as Embeddings
|
|
const basePath = nodeData.inputs?.basePath as string
|
|
|
|
const flattenDocs = docs && docs.length ? flatten(docs) : []
|
|
const finalDocs = []
|
|
for (let i = 0; i < flattenDocs.length; i += 1) {
|
|
if (flattenDocs[i] && flattenDocs[i].pageContent) {
|
|
finalDocs.push(new Document(flattenDocs[i]))
|
|
}
|
|
}
|
|
|
|
try {
|
|
const vectorStore = await FaissStore.fromDocuments(finalDocs, embeddings)
|
|
await vectorStore.save(basePath)
|
|
|
|
// Avoid illegal invocation error
|
|
vectorStore.similaritySearchVectorWithScore = async (query: number[], k: number) => {
|
|
return await similaritySearchVectorWithScore(query, k, vectorStore)
|
|
}
|
|
|
|
return { numAdded: finalDocs.length, addedDocs: finalDocs }
|
|
} catch (e) {
|
|
throw new Error(e)
|
|
}
|
|
}
|
|
}
|
|
|
|
async init(nodeData: INodeData): Promise<any> {
|
|
const embeddings = nodeData.inputs?.embeddings as Embeddings
|
|
const basePath = nodeData.inputs?.basePath as string
|
|
const output = nodeData.outputs?.output as string
|
|
const topK = nodeData.inputs?.topK as string
|
|
const k = topK ? parseFloat(topK) : 4
|
|
|
|
const vectorStore = await FaissStore.load(basePath, embeddings)
|
|
|
|
// Avoid illegal invocation error
|
|
vectorStore.similaritySearchVectorWithScore = async (query: number[], k: number) => {
|
|
return await similaritySearchVectorWithScore(query, k, vectorStore)
|
|
}
|
|
|
|
if (output === 'retriever') {
|
|
const retriever = vectorStore.asRetriever(k)
|
|
return retriever
|
|
} else if (output === 'vectorStore') {
|
|
;(vectorStore as any).k = k
|
|
return vectorStore
|
|
}
|
|
return vectorStore
|
|
}
|
|
}
|
|
|
|
const similaritySearchVectorWithScore = async (query: number[], k: number, vectorStore: FaissStore) => {
|
|
const index = vectorStore.index
|
|
|
|
if (k > index.ntotal()) {
|
|
const total = index.ntotal()
|
|
console.warn(`k (${k}) is greater than the number of elements in the index (${total}), setting k to ${total}`)
|
|
k = total
|
|
}
|
|
|
|
const result = index.search(query, k)
|
|
return result.labels.map((id, index) => {
|
|
const uuid = vectorStore._mapping[id]
|
|
return [vectorStore.docstore.search(uuid), result.distances[index]] as [Document, number]
|
|
})
|
|
}
|
|
|
|
module.exports = { nodeClass: Faiss_VectorStores }
|