Adamdad/DeRy

[NeurIPS2022] Deep Model Reassembly

27
/ 100
Experimental

This project helps machine learning engineers and researchers create customized neural networks from existing, pre-trained models. It takes a collection of trained deep learning models as input, dissects them into their fundamental building blocks, and then intelligently reassembles these blocks to construct new, specialized networks. The output is a highly optimized model tailored for specific tasks, adhering to hardware resource and performance constraints, without needing to train a new model from scratch.

253 stars. No commits in the last 6 months.

Use this if you need to quickly build a new, efficient deep learning model for a specific task by reusing components from a library of existing models, especially when facing hardware resource limitations or performance targets.

Not ideal if you prefer to design and train all components of your deep learning models from the ground up, or if your application requires highly novel architectures that cannot be effectively formed by reassembling existing blocks.

model-optimization neural-architecture-search transfer-learning resource-constrained-ai deep-learning-engineering
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 8 / 25
Community 9 / 25

How are scores calculated?

Stars

253

Forks

11

Language

Python

License

Last pushed

Aug 29, 2023

Commits (30d)

0

Get this data via API

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

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