ironjr/grokfast

Official repository for the paper "Grokfast: Accelerated Grokking by Amplifying Slow Gradients"

42
/ 100
Emerging

This tool helps machine learning researchers and practitioners accelerate the "grokking" phenomenon in their models. When training deep learning models, sometimes they overfit to the training data perfectly before suddenly generalizing well to new, unseen data, which is called grokking. This tool takes your existing PyTorch model's gradients and processes them to drastically speed up this generalization process, potentially reducing training time significantly.

578 stars. No commits in the last 6 months.

Use this if your deep learning model exhibits grokking, where it overfits the training data for a long time before finally generalizing, and you want to speed up that generalization phase.

Not ideal if your model does not exhibit the grokking phenomenon or if you are not working with PyTorch models.

deep-learning-research model-training neural-network-optimization gradient-descent machine-learning-engineering
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 16 / 25

How are scores calculated?

Stars

578

Forks

50

Language

Python

License

MIT

Last pushed

Jun 28, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/ironjr/grokfast"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.