snimu/rebasin

Apply methods described in "Git Re-basin"-paper [1] to arbitrary models --- [1] Ainsworth et al. (https://arxiv.org/abs/2209.04836)

46
/ 100
Emerging

This tool helps machine learning engineers and researchers combine multiple trained neural network models, or understand the 'loss landscape' between them. You provide two or more pre-trained models, and it can 'rebasin' them to align their internal representations or interpolate smoothly between their weights. This process yields a new, potentially better-performing merged model or a sequence of interpolated models for analysis.

Available on PyPI.

Use this if you need to merge several independently trained neural networks into a single, stronger model, or if you want to analyze the 'path' between two models in terms of their performance.

Not ideal if your models are primarily transformer-based architectures, as they often don't benefit well from the permutation techniques implemented here.

deep-learning model-merging neural-network-analysis computer-vision model-optimization
Maintenance 10 / 25
Adoption 6 / 25
Maturity 25 / 25
Community 5 / 25

How are scores calculated?

Stars

15

Forks

1

Language

Python

License

MIT

Last pushed

Mar 09, 2026

Commits (30d)

0

Dependencies

6

Get this data via API

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

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