अपस्टैश के साथ एक शक्तिशाली RAG चैटबॉट बनाएं:एक चरण-दर-चरण मार्गदर्शिका

आवश्यकताएँ

• Node.js 18 या बाद का संस्करण
• एक अपस्टैश खाता
• ए हगिंग फेस अकाउंट
• एक प्रतिकृति खाता
• एक वर्सेल खाता

टेक स्टैक

अपस्टैश रेडिस की स्थापना

अपस्टैश वेक्टर सेट करना

प्रोजेक्ट की स्थापना

git clone https://github.com/rishi-raj-jain/custom-rag-chatbot-upstash-vector
cd custom-rag-chatbot-upstash-vector
pnpm install

# .env
 
# Obtained from the steps as above
 
# Upstash Redis URL and Token
UPSTASH_REDIS_REST_URL="https://....upstash.io"
UPSTASH_REDIS_REST_TOKEN="..."
 
# Upstash Vector URL and Token
UPSTASH_VECTOR_REST_URL="https://...-vector.upstash.io"
UPSTASH_VECTOR_REST_TOKEN="..."
 
# Replicate API Key
REPLICATE_API_TOKEN="r8_..."
 
# Hugging Face Inference API Key
HUGGINGFACEHUB_API_KEY="hf_..."

pnpm dev

भंडार संरचना

नेक्स्ट.जेएस ऐप राउटर में चैट रूट सेटअप करें

बातचीत संग्रहीत करना

// File: app/api/chat/route.js
 
import { Redis } from '@upstash/redis'
 
// Instantiate the Upstash Redis
const upstashRedis = new Redis({
 url: process.env.UPSTASH_REDIS_REST_URL,
 token: process.env.UPSTASH_REDIS_REST_TOKEN,
})
 
export async function POST(req) {
 try {
 
 // the whole chat as array of messages
 const { messages } = await req.json()
 
 // assuming user - assistant chat
 // add assitant's response to the chat history
 if (messages.length > 1) {
 await upstashRedis.lpush('unique_conversation_id', JSON.stringify(messages[messages.length - 2]))
 }
 // add user's request to the chat history
 await upstashRedis.lpush('unique_conversation_id', JSON.stringify(messages[messages.length - 1]))
 
 // Proceed to create a response
}

नवीनतम संदेश की एम्बेडिंग बनाएं

// File: app/api/chat/route.js
 
import { HuggingFaceInferenceEmbeddings } from '@langchain/community/embeddings/hf'
 
// Instantiate the Hugging Face Inference API
const embeddings = new HuggingFaceInferenceEmbeddings()
 
export async function POST(req) {
 try {
 // ...
 
 // get the latest question stored in the last message of the chat array
 const userMessages = messages.filter((i) => i.role === 'user')
 const lastMessage = userMessages[userMessages.length - 1].content
 
 // generate embeddings of the latest question
 const queryVector = (await embeddings.embedQuery(lastMessage)).slice(0, 256)
 
 // Proceed to create a response
}

नवीनतम संदेश के आधार पर प्रासंगिक संदर्भ वैक्टर पुनर्प्राप्त करें

// File: app/api/chat/route.js
 
import { Index } from '@upstash/vector'
 
// Instantiate the Upstash Vector Index
const upstashVectorIndex = new Index()
 
export async function POST(req) {
 try {
 // ...
 
 // query the relevant vectors from the embedding vector
 const queryResult = await upstashVectorIndex.query({
 vector: queryVector,
 // get the top 2 relevant results
 topK: 2,
 // do not include the whole set of embeddings in the response
 includeVectors: false,
 // include the meta data so that can get the description out of the index
 includeMetadata: true,
 })
 // console.log('The query result came in', queryResult.length)
 
 // using the resulting set of relevant vectors
 // filter the one that have score of greater than 70% match
 // and get the description we stored while training
 const queryPrompt = queryResult
 .filter((match) => match.score && match.score > 0.7)
 .map((match) => match.metadata.description)
 .join('\n')
 // console.log('The query prompt is', queryPrompt)
 
 // Proceed to create a response
}

भविष्यवाणियों के संदर्भ के साथ LLAMA-2-70B चैट मॉडल का संकेत दें

// File: app/api/chat/route.js
 
import Replicate from 'replicate'
import { experimental_buildLlama2Prompt } from 'ai/prompts'
import { ReplicateStream, StreamingTextResponse } from 'ai'
 
// Instantiate the Replicate API
const replicate = new Replicate({
 auth: process.env.REPLICATE_API_TOKEN,
})
 
export async function POST(req) {
 try {
 // ...
 
 const response = await replicate.predictions.create({
 // You must enable streaming.
 stream: true,
 // The model must support streaming. See https://replicate.com/docs/streaming
 // This is the model ID for Llama 2 70b Chat
 version: '2c1608e18606fad2812020dc541930f2d0495ce32eee50074220b87300bc16e1',
 // Format the message list into the format expected by Llama 2
 // @see https://github.com/vercel/ai/blob/99cf16edf0a09405d15d3867f997c96a8da869c6/packages/core/prompts/huggingface.ts#L53C1-L78C2
 input: {
 prompt: experimental_buildLlama2Prompt([
 {
 // create a system content message to be added as
 // the llama2prompt generator will supply it as the context with the API
 role: 'system',
 content: queryPrompt.substring(0, Math.min(queryPrompt.length, 2000)),
 },
 // also, pass the whole conversation!
 ...messages,
 ]),
 },
 })
 
 // stream the result to the frontend
 const stream = await ReplicateStream(response)
 return new StreamingTextResponse(stream)
}

नेक्स्ट.जेएस ऐप राउटर में ट्रेन रूट सेटअप करें

स्ट्रिंग्स की एम्बेडिंग बनाएं

// File: app/api/train/route.js
 
import { HuggingFaceInferenceEmbeddings } from '@langchain/community/embeddings/hf'
 
// Instantiate the Hugging Face Inference API
const embeddings = new HuggingFaceInferenceEmbeddings()
 
export async function POST(req) {
 try {
 
 // a default set of messages to create vector embeddings on
 let messagesToVectorize = [
 'Rishi is pretty much active on Twitter nowadays.',
 'Rishi loves writing for Upstash',
 "Rishi's recent article on building chatbot using Upstash went viral",
 'Rishi is enjoying building launchfa.st.',
 ]
 
 // if the POST request is of type application/json
 if (req.headers.get('Content-Type') === 'application/json') {
 // and if the request contains array of messages to train on
 const { messages } = await req.json()
 if (typeof messages !== 'string' && messages.length > 0) {
 messagesToVectorize = messages
 }
 }
 
 // Call the Hugging Face Inference API to get emebeddings on the messages
 const generatedEmbeddings = await Promise.all(messagesToVectorize.map((i) => embeddings.embedQuery(i)))
 
 // ...
}

प्रासंगिकता खोज के लिए वैक्टर संग्रहीत करें

// File: app/api/train/route.js
 
import { Index } from '@upstash/vector'
 
// Instantiate the Upstash Vector Index
const upstashVectorIndex = new Index()
 
export async function POST(req) {
 try {
 // ...
 
 // Slice the vector into lengths of upto 256
 await Promise.all(
 generatedEmbeddings
 .map((i) => i.slice(0, 256))
 .map((vector, index) =>
 // Upsert the vector with description to be further as the context to upcoming questions
 upstashVectorIndex.upsert({
 vector,
 id: index.toString(),
 metadata: { description: messagesToVectorize[index] },
 }),
 ),
 )
 // Once done, return with a successful 200 response
 return new Response(JSON.stringify({ code: 1 }), { status: 200, headers: { 'Content-Type': 'application/json' } })
}

वर्सेल में तैनात करें

• अपने ऐप के कोड वाली GitHub रिपॉजिटरी बनाकर शुरुआत करें।
• फिर, वर्सेल डैशबोर्ड पर नेविगेट करें और एक नया प्रोजेक्ट बनाएं .
• नए प्रोजेक्ट को आपके द्वारा अभी बनाए गए GitHub रिपॉजिटरी से लिंक करें।
• सेटिंग्स में , Environment Variables को अपडेट करें  आपके स्थानीय .env से मिलान करने के लिए  फ़ाइल.
• तैनाती! 🚀

अधिक जानकारी

निष्कर्ष