This repository contains a Python script that demonstrates how to use transformer models to convert code snippets into vector embeddings. These embeddings can then be used to perform code search using natural language queries.
load_codebase: Recursively navigates through a directory, filtering out unwanted files and directories, and loads the content of the allowed files into a list of code snippets.average_pool: Averages out the token embeddings to get a fixed-size vector for each code snippet.generate_embeddings: Converts code snippets into vector representations using a pre-trained transformer model.find_k_nearest_neighbors: Calculates the cosine similarity between a query embedding and all code snippet embeddings to find the most relevant pieces of code.
- Clone this repository.
- Ensure you have the required libraries installed.
- Run the script to see an example of how the code search works.
python app.pyFor a step-by-step walkthrough of the code and the concepts behind it, check out the accompanying blog post: Code Search with Vector Embeddings: A Transformer's Approach.