LangChain playground

Pre-requisites

• Docker.
• An Elasticsearch cluster running 8.9 or greater.
• A working Python 3.11 or greater installation.
• An OpenAI API key to use LLM integrations.

Setup

• Go the directory where you cloned the repository.

• Create a virtual env

  python3 -mvenv .venv

• Activate the virtual env (source .venv/bin/activate for bash/zsh, .venv\bin\activate on Windows)

• Install poetry

  pip install -U pip wheel poetry

• Install dependencies:

  poetry install

Basic configuration

• Copy the file config/sample.yaml to config/default.yaml .
• Fill the elasticsearch stanza with the URL and credentials to access your Elasticsearch cluster.
• Set elasticsearch.index to the index where you will store data. ⚠️ Make sure to not put an existing index here as it will be destroyed.

Indexing

Put at least one PDF in the docs folder and run python index.py --nuke to remove existing documents and start processing.

On the first execution the script will install any required dependency / model from HuggingFace.

Search

Start the API with python serve.py.

You can visit the API docs at http://localhost:8000 and send a question through the ask endpoint; if indexing went well, you should get results.

LLM integration

To enable prompting the question to an LLM, you will need to set the following options:

• openai.apikey: your OpenAI API key.
• openai.completionModel: the completion model to use (gpt-3.5-turbo-16k by default).
• llm: set to openai.

After restarting the server, requests to ask should return the final response from the LLM using the results as a context.

UI

A simple streamlit app to use the API is available by running streamlit run ui.py or docker-compose up -d .

Once started, it can be viewed at http://localhost:8501 .