Ekoda/SoftMoE

Soft Mixture of Experts Vision Transformer, addressing MoE limitations as highlighted by Puigcerver et al., 2023.

23
/ 100
Experimental

This project provides an implementation of a Soft Mixture of Experts Vision Transformer, designed for developers working with large-scale deep learning models. It takes in model configurations and image data, producing a trained or fine-tuned vision transformer model that avoids common issues like token dropping and training instability found in traditional Mixture of Experts (MoE) architectures. This is for machine learning engineers and researchers building or experimenting with vision models.

No commits in the last 6 months.

Use this if you are developing computer vision models and want to leverage Mixture of Experts architectures while avoiding the common training and scaling limitations of sparse MoE models.

Not ideal if you are looking for a pre-trained model for immediate use or are not comfortable working with deep learning model implementations at a code level.

deep-learning computer-vision machine-learning-engineering model-architecture neural-networks
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 6 / 25
Maturity 8 / 25
Community 9 / 25

How are scores calculated?

Stars

16

Forks

2

Language

Python

License

Last pushed

Aug 13, 2023

Commits (30d)

0

Get this data via API

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

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