Show HN: Graph embedding based code suggestion framework

1 point

a year ago

Introducing GraphSense GraphSense is a framework that can be used to easily train and use code suggestion models with minimal data preprocessing and resource consumption. No transformers were used and underlying algorithm used was Node2Vec. Models trained will be able to suggest next code line using current code line.

locally training GPT models for code suggestions takes too much time and resources

GPT-based code suggestions take too long and consume more memory

But GraphSense takes code suggestions down a different path by ditching transformers for more efficient graph embedding-based code suggestions.

GPT2-medium model memory usage: 1800 MB GPT2-medium model inference time (GPU): 2 seconds

GraphSense memory usage (1M vocabulary): 734 MB GraphSense inference time: 0.0444 seconds

Install using: pip install graphsense Github Repo: https://github.com/NavodPeiris/graphsense

If you like it, start it