Skip to content

Latest commit

 

History

History
165 lines (99 loc) · 5.53 KB

README.md

File metadata and controls

165 lines (99 loc) · 5.53 KB
Raw

This Repo is for creating RAG system using LangChain and Streamlit

Python Environment

1. Install Packages

pip install -r requirements.txt

2. Set Api Key

create_api_key

  • Copy it into .env file

​ Set OPENAI_API_KEY="Your API KEY"

3. Run Simple Version On Colab (only support one pdf file)

  • Import colab.ipynb into Google Colab.

  • Drag your pdf file into Google Colab and change the file name in the code.

loader = PyPDFLoader("data.pdf")
  • Input your openai api key in the ChatOpenAI().
llm = ChatOpenAI(model_name="gpt-3.5-turbo", openai_api_key="")
  • You can change embedding model by searching on HuggingFace.
embedding = HuggingFaceEmbeddings(model_name="sentence-transformers/xxxxxxx")
  • Ask question and get answer on Google Colab.

4. Run Streamlit On Colab

  • Import localtunnel.ipynb into Google Colab.

  • Input your openai api key in the ChatOpenAI().

llm = ChatOpenAI(
    model_name="gpt-3.5-turbo",
    temperature=0.1,
    openai_api_key=""
)
  • You can change embedding model by searching on HuggingFace.
embedding = HuggingFaceEmbeddings(
    model_name="sentence-transformers/xxxxxxx",
    model_kwargs={"device": "cpu"}
)
  • You can get three urls, but you don't need to click any of them, stop this cell.

npx url

  • Run the next cell, get the tunnel password.

get curl password

  • Run the above cell again, you can see three urls. Click the last url and you will see the web page below.

password UI

  • Enter the tunnel password, which you got in the previous step. Then you can see the Streamlit WebUI.

streamlit ui

5. Run Streamlit On Local Computer

streamlit run app.py

After running this command, you can see the WebUI as the image above. On the left side, you can choose "Browse files" to upload multiple files as long as they are pdf, doc or txt format. If you encounter the error AxiosError: Request failed with status code 403 while uploading the file. Try the command below.

streamlit run app.py --server.enableXsrfProtection false

Then you should be able to upload files successfully, like the image below.

You need to wait for some time to let the embedding model convert all files into high dimensional vectors and store them into a database. You will see a new folder in your local computer.

Then you can see the original chat interface like this.

Feel free to enter your question and click Send to chat with the RAG system, you may need to wait for about 15 seconds after sending your new query.

6. Compare RAG With Original ChatGPT

python compare.py

The code is almost the same as colab.ipynb, just add the response from original ChatGPT. When you enter the question, you can see responses from both RAG system and original ChatGPT.

7. Deploy Your App

  • Fork this GitHub repo into your own GitHub account

fork

  • Set your OPENAI_API_KEY in the .env file. (You need to clone the repo to local computer, change the file and commit it, or maybe you can delete this file and upload an another .env file)

set key

create new app

  • Enter your GitHub Repo Url in Repository and change the Main file path to app.py

add repo url

  • Click Deploy!, wait for installing all packages in the requirements.txt, you can see the progress.

deploy process

  • After it's done, you can use it.

chat

see your app

8. Use Groq For Faster Inference

  • Get your GROQ_API_KEY at https://console.groq.com/keys.

  • Set your GROQ_API_KEY in the .env file.

  • Follow all the steps in the Part 7, but change the Main file path to app-groq.py

9. Use Llama Model

10. Warning

Since others will be using your OPENAI_API_KEY, GROQ_API_KEY and LLAMA_API_KEY when you share your app with them, all costs will be charged to your account. Be cautious when sharing it, and if needed, promptly delete the application you've deployed.