Compare commits

..

No commits in common. "c3dac38b6a0bcdbc801a73580953dfd6148c8ccf" and "df8d924a89c5373855a3d78ec0a37711a1774d85" have entirely different histories.

3 changed files with 30 additions and 41 deletions

View File

@ -157,43 +157,35 @@ const createBasicWebSearchRetrieverChain = (llm: BaseChatModel) => {
question = 'Summarize';
}
let docs = [];
let docs = []
const linkDocs = await getDocumentsFromLinks({ links });
const docGroups: Document[] = [];
linkDocs.map((doc) => {
const URLDocExists = docGroups.find(
(d) =>
d.metadata.url === doc.metadata.url && d.metadata.totalDocs < 10,
);
const URLDocExists = docGroups.find((d) => d.metadata.url === doc.metadata.url && d.metadata.totalDocs < 10);
if (!URLDocExists) {
docGroups.push({
...doc,
metadata: {
...doc.metadata,
totalDocs: 1,
},
totalDocs: 1
}
});
}
const docIndex = docGroups.findIndex(
(d) =>
d.metadata.url === doc.metadata.url && d.metadata.totalDocs < 10,
);
const docIndex = docGroups.findIndex((d) => d.metadata.url === doc.metadata.url && d.metadata.totalDocs < 10);
if (docIndex !== -1) {
docGroups[docIndex].pageContent =
docGroups[docIndex].pageContent + `\n\n` + doc.pageContent;
docGroups[docIndex].pageContent = docGroups[docIndex].pageContent + `\n\n` + doc.pageContent;
docGroups[docIndex].metadata.totalDocs += 1;
}
});
})
await Promise.all(
docGroups.map(async (doc) => {
const res = await llm.invoke(`
await Promise.all(docGroups.map(async (doc) => {
const res = await llm.invoke(`
You are a text summarizer. You need to summarize the text provided inside the \`text\` XML block.
You need to summarize the text into 1 or 2 sentences capturing the main idea of the text.
You need to make sure that you don't miss any point while summarizing the text.
@ -212,17 +204,16 @@ const createBasicWebSearchRetrieverChain = (llm: BaseChatModel) => {
Make sure to answer the query in the summary.
`);
const document = new Document({
pageContent: res.content as string,
metadata: {
title: doc.metadata.title,
url: doc.metadata.url,
},
});
const document = new Document({
pageContent: res.content as string,
metadata: {
title: doc.metadata.title,
url: doc.metadata.url,
},
})
docs.push(document);
}),
);
docs.push(document)
}))
return { query: question, docs: docs };
} else {

View File

@ -30,9 +30,9 @@ server.listen(port, () => {
startWebSocketServer(server);
process.on('uncaughtException', (err, origin) => {
logger.error(`Uncaught Exception at ${origin}: ${err}`);
});
logger.error(`Uncaught Exception at ${origin}: ${err}`)
})
process.on('unhandledRejection', (reason, promise) => {
logger.error(`Unhandled Rejection at: ${promise}, reason: ${reason}`);
});
logger.error(`Unhandled Rejection at: ${promise}, reason: ${reason}`)
})

View File

@ -1,8 +1,8 @@
import axios from 'axios';
import { htmlToText } from 'html-to-text';
import { htmlToText } from 'html-to-text'
import { RecursiveCharacterTextSplitter } from 'langchain/text_splitter';
import { Document } from '@langchain/core/documents';
import pdfParse from 'pdf-parse';
import pdfParse from 'pdf-parse'
export const getDocumentsFromLinks = async ({ links }: { links: string[] }) => {
const splitter = new RecursiveCharacterTextSplitter();
@ -23,14 +23,14 @@ export const getDocumentsFromLinks = async ({ links }: { links: string[] }) => {
const isPdf = res.headers['content-type'] === 'application/pdf';
if (isPdf) {
const pdfText = await pdfParse(res.data);
const pdfText = await pdfParse(res.data)
const parsedText = pdfText.text
.replace(/(\r\n|\n|\r)/gm, ' ')
.replace(/\s+/g, ' ')
.trim();
const splittedText = await splitter.splitText(parsedText);
const title = 'PDF Document';
const title = 'PDF Document'
const linkDocs = splittedText.map((text) => {
return new Document({
@ -52,18 +52,16 @@ export const getDocumentsFromLinks = async ({ links }: { links: string[] }) => {
selector: 'a',
options: {
ignoreHref: true,
},
}
},
],
]
})
.replace(/(\r\n|\n|\r)/gm, ' ')
.replace(/\s+/g, ' ')
.trim();
const splittedText = await splitter.splitText(parsedText);
const title = res.data
.toString('utf8')
.match(/<title>(.*?)<\/title>/)?.[1];
const title = res.data.toString('utf8').match(/<title>(.*?)<\/title>/)?.[1];
const linkDocs = splittedText.map((text) => {
return new Document({