README.md

Table of Contents

• Prerequisites
• Foreword
• Project setup
• Debugging
  • Alternative ingest with python
  • Openai paid account
  • Openai free trial expired
  • Issues with GPT4 api key
  • Check if pinecone is working
• Contribution

Forwords to China

Some have reported issues with this project when they start it in china. There might be regional blocking from openai/pinecone (21.05.2023).

Solution from jchermy

1. : Create Cloudflare Worker, refer to noobnooc/noobnooc#9;
2. ：Proxy all the needs of openai. The example is as follows:

const embeddings = new OpenAIEmbeddings(undefined, {
  basePath: 'https://openai.1rmb.tk/v1/',
  apiKey: 'xxx',
}); 

Prerequisites

• Openai account with payment option.
  • Check in this section, if your openai account is appropiate: issues-with-openai, [Openai free trial expired](https://github.com/ucl98/gpt4_langchain_pdf-detailed-walktrough#Openai free trial expired).
    • Having a openai membership for ChatGPT is NOT sufficient.
  • As rough estimate: 1000pdf pages cost about 1€.
  • Use GPT3.5 key first before trying GPT4 key (Issues with GPT4 api key).
    • GPT4 key may cause issues or will be less performaned.
• Pinecone account.

Discord Server

Join the discord if you have questions.

Project setup

Download the project

Download the project: https://github.com/mayooear/gpt4-pdf-chatbot-langchain

Install nodejs and npm

Install Node.js and npm. Official website: https://nodejs.org/en/download

Verfiy that everything is installed with the following terminal commands.

• Your version may differ, it is only important that node has version 18 or higher.

Install all required packages

Open a terminal at the top level of the project and run "npm install" to install all necessary dependencies.

Create a ".env"

Create a file named ".env" inside your project with the following content:

OPENAI_API_KEY = 

PINECONE_API_KEY = 
PINECONE_ENVIRONMENT = 

PINECONE_INDEX_NAME = 

Openai api key

Go to https://platform.openai.com/overview and got "View API keys".

Click "+ Create new secrete key" and copy the key in your ".env" file

Pinecone api key

Go to https://app.pinecone.io/organizations/ and create an api key there as well. Add them to the ".env" file.

Pinecone index

Create a new pinecone index with the Dimension 1536.

It should look similar to this:

Add pinecone index to the ".env" file

Add the information from your index correct to the .env file. At this point it should look like this:

(Optional) Set namespace in config/pinecone.ts

Give the namespace for your vectors a name in "config/pinecone.ts".

All vectors that are ingested will be available by this namespace in your pinecone index. The name you defined in "config/pinecone.ts" will be used to store your ingested data in this namespace, but also lookup your data on a question to the chatbot.

• It is not possible to search data in multiple namespaces simultaneously.

Docs folder

Create a folder "docs" where you have your pdf files.

Ingest your pdf

Run "npm run ingest" in a terminal. This will ingest your pdf into the pinecone database. Remove the pdf document from the "docs" folder, otherwise it will be ingested a second time if you run "npm run ingest" again.

The terminal must be in the project folder at the top level. I have installed "gpt4-pdf-chatbot-langchain" in "Documents".

Start appliation

Start the application with "npm run dev" in the terminal. Open a browser and go to the address http://localhost:3000/. You should see the project.

Debugging

Alternative ingest with python

To trouble shoot it might be useful to take a look at my python reimplementation at ingest. It gives more control over each step of ingest.

https://github.com/ucl98/pinecone_ingest_python_implementation

Openai paid account

Check if you have a paid account. If you see the following message, you need to set up a paid account.

• https://platform.openai.com/account/usage

Openai free trial expired

You may used a free trial for the openai api. This trial expires and in this case you cannot use the openai api. For this case add a payed option.

Issues with GPT4 api key

Go to openai playgroud and check if you have access to GPT4. If the model is not available, you will have no access to it and need to use GPT3.5.

• Access to the GPT4 api key does not depend on a ChatGPT subscription. You needd to apply for it.

Check if pinecone is working

You can make a pinecone api request on the webpage of pinecone. Go to the "Query" tab, enter the namespace, approve the namespace and then hit "query". If it returns a result, the connection to pinecone is working.

Api keys

Another issue can be that your api key is not working. Create a new one and test it again.

Contribution

• ucl
• angelina-magidova-synder
• chaudhary_181
• jchermy