samuela/git-re-basin
Code release for "Git Re-Basin: Merging Models modulo Permutation Symmetries"
This project helps machine learning researchers understand and compare independently trained neural networks. It takes two trained models and applies a permutation matching algorithm to their internal layers. The output helps analyze if their performance differences are due to fundamental landscape variations or just different internal arrangements of neurons. This is useful for deep learning researchers and practitioners exploring model convergence and loss landscapes.
509 stars. No commits in the last 6 months.
Use this if you need to determine whether two neural networks, trained independently, occupy fundamentally similar areas of the loss landscape after accounting for internal neuron arrangements.
Not ideal if you are looking to directly improve model performance or merge models for deployment without deep analysis of their internal structures.
Stars
509
Forks
44
Language
Python
License
MIT
Category
Last pushed
Mar 07, 2023
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/samuela/git-re-basin"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
zeiss-microscopy/OAD
Collection of tools and scripts useful to automate microscopy workflows in ZEN Blue using Python...
median-research-group/LibMTL
A PyTorch Library for Multi-Task Learning
qdrant/quaterion
Blazing fast framework for fine-tuning similarity learning models
lucidrains/PoPE-pytorch
Efficient implementation (and explorations) into polar coordinate positional embedding (PoPE) -...
MR-HosseinzadehTaher/BenchmarkTransferLearning
Official PyTorch Implementation and Pre-trained Models for Benchmarking Transfer Learning for...