Skip to content

Latest commit

 

History

History
65 lines (52 loc) · 2.68 KB

README.md

File metadata and controls

65 lines (52 loc) · 2.68 KB
Raw

nush-llm-chatbot

HOWTO

Running of the prototype webpage requires the setup of a python environment and the construction of the vector database.

Step 0: API KEY Create a file called "apikey" under /llamaindex directory. The only contents in the file should be the openai api key. The apikey is also present under the teams development folder in the same /llamaindex directory, copying the file over also works.

Step 1: Setup python environment Environment package managers like pipenv or anaconda are recommended, these environment managers ensure that isolated python workspaces are created and packages installed do not interfere with the packages of another environment. There are many tutorials online on how to download and setup a package environment.

The requirements.txt file required to download relevant packages has already been provided. Navigate to the root directory and run this command (ensure the correct python environment is active):

pip install -r requirements.txt

Now just have to wait for the packages to complete downloading.

Step 2: Setup vector database This step only has to be done once, and caches all the vectors in the folder /llamaindex/storage/ for the model to refer to. There are two options, one is to run the file and the other is to copy the already created file from teams.

Simply run /llamaindex/vector_indices.py via the editor or through the command line as shown:

python llamaindex/vector_indices.py

This should create a storage folder if it does not already exist. Alternatively, the file /llamaindex/storage/ from Teams can be copied over and placed in the same directory locally.

Step 3: Run app.py either through the editor or through command line console:

python -m frontend.app

The output should display something like:

 * Serving Flask app 'app'
 * Debug mode: on
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
 * Running on all addresses (0.0.0.0)
 * Running on http://127.0.0.1:5601
 * Running on http://172.20.10.10:5601
Press CTRL+C to quit

The IP address that the server is running on might vary, but opening the displayed address in a web browser will open up the app. Once the webpage is opened, the first time executing the application might be met with some buffer period as the huggingface model might have to be downloaded and initialised for the first time.

TODO