davda54/sam
A PyTorch optimizer wrapper implementing Sharpness-Aware Minimization to improve deep learning model generalization.
Not currently ranked — collecting fresh signals.
star history
This repository provides an unofficial implementation of SAM and ASAM optimizers for PyTorch. SAM works by performing two forward-backward passes to compute a regularized sharpness-aware gradient, which is then passed to an underlying optimizer like SGD with momentum. The technique leads to wider loss minima and improved generalization across various datasets, with robustness to label noise comparable to specialized noisy-label training procedures.
Frequently asked
- What is davda54/sam?
- A PyTorch optimizer wrapper implementing Sharpness-Aware Minimization to improve deep learning model generalization.
- Is sam open source?
- Yes — davda54/sam is open source, released under the MIT license.
- What language is sam written in?
- davda54/sam is primarily written in Python.
- How popular is sam?
- davda54/sam has 2k stars on GitHub.
- Where can I find sam?
- davda54/sam is on GitHub at https://github.com/davda54/sam.