omar-florez / llm_online

Augment LLM reasoning with results from a search engine.

Description

LLMs learn static representation of knowledge in the weigths. Our experiments show that smaller models (7B) can compete with models of higher capacity such as ChatGPT and Llama 2 in perceived utility and F1 score if they access search engines to augment their reasoning power during in-context learning.

Steps to run

1. Create a file named .env in the root folder and define your API keys for OpenAI and SearchAPI.

• touch .env

openai_api_key = ''
serpapi_api_key = ''

2. Run with:

• pip install -r requirements.txt
• python run/run_online.py