hkchengrex/nitrous-ema
Fast and simple post-hoc EMA (Karras et al., 2023) for PyTorch with minimal `.item()` calls. ~78% lower overhead than ema_pytorch.
This tool helps machine learning engineers efficiently apply Exponential Moving Average (EMA) to their PyTorch models after training. It takes your trained model's weights and configuration parameters for EMA, then produces an EMA-smoothed version of your model that often performs better. This is for machine learning practitioners and researchers working with deep learning models in PyTorch.
No commits in the last 6 months. Available on PyPI.
Use this if you need a significantly faster way to apply post-hoc EMA to your PyTorch models, especially when training large neural networks where synchronization overhead matters.
Not ideal if you are not using PyTorch or if you prefer a different method for model stabilization beyond post-hoc EMA.
Stars
13
Forks
—
Language
Python
License
MIT
Category
Last pushed
Nov 02, 2024
Commits (30d)
0
Dependencies
1
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/hkchengrex/nitrous-ema"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
pytorch/pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
keras-team/keras
Deep Learning for humans
Lightning-AI/torchmetrics
Machine learning metrics for distributed, scalable PyTorch applications.
Lightning-AI/pytorch-lightning
Pretrain, finetune ANY AI model of ANY size on 1 or 10,000+ GPUs with zero code changes.
lanpa/tensorboardX
tensorboard for pytorch (and chainer, mxnet, numpy, ...)