chenjoya/dropit

DropIT: Dropping Intermediate Tensors for Memory-Efficient DNN Training (ICLR 2023)

24
/ 100
Experimental

This project helps machine learning engineers train large deep neural networks more efficiently by reducing the GPU memory required. It takes an existing neural network model and applies a technique that prunes less important information from intermediate calculations. The outcome is a model that trains faster or can handle larger batch sizes on the same hardware, often achieving even better accuracy for tasks like image classification, object detection, and instance segmentation.

No commits in the last 6 months.

Use this if you are training deep learning models, especially large ones like Vision Transformers or Convolutional Neural Networks, and are frequently running into GPU memory limits or want to reduce training time.

Not ideal if your models are small and already train very quickly without memory issues, or if you require absolute precision in every intermediate tensor calculation.

deep-learning-training computer-vision natural-language-processing gpu-optimization neural-network-efficiency
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 9 / 25

How are scores calculated?

Stars

32

Forks

3

Language

Python

License

Last pushed

Apr 08, 2023

Commits (30d)

0

Get this data via API

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

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