Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A modern deep learning framework built to accelerate research and development of AI systems
| Date | Stars |
|---|---|
| 2026-07-31 | 258 |
| 2026-08-06 | 258 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# TorchFusion
A modern deep learning framework built to accelerate research and development of AI systems.
Based on PyTorch and fully compatible with pure PyTorch and other pytorch packages, <b>TorchFusion</b> provides a comprehensive extensible training framework
with trainers that you can easily use to train, evaluate and run inference with your PyTorch models, A GAN framework that greatly simplifies the process of
experimenting with Generative Adversarial Networks [Goodfellow et al. 2014](https://arxiv.org/1406.2661), with concrete implementations of a number of GAN algorithms, and a number of high level network layers and utilities to help you be more productive in your work.
<p>The framework is highly extensible, so you can easily create your own custom trainers for specific purposes.</p>
# New in 2.0
* Improved Trainer Framework
* Support for multiple Inputs and Outputs
* New utilities for loading images, one-hot encoding and more.
* New Gan Framework with multiple layers of abstraction and implementation of
Hinge GANs, GANs with divergence loss, Wasserstein GANs and Relativistic GANs.
* New GAN Applications with support for spectral normalization, conditional batch normalization, self attention, projection gans and resnet generators and discriminators
* A wider range of Initializers
* Enhanced summary function that not only provides you details about number of parameters, layers, input and output sizes
but also provides the number of Flops(Multiply-Adds) for every Linear and Convolution layer in your network.
Now, you can know the exact computational cost of any CNN architecure with just a single function!!!
* Visdom and Tensorboard Support
* Live metrics and loss visualizations, with option to save them permanently
* Support for persisting logs permanently
* Easy to use callbacks
<b>Note: This version of torchfusion is well tested and research-ready, the core framework is now complete, Future releases of TorchFusion will include more specialized functions that will cut across multiple domains of deep learning
</b>
An <b>AI Commons</b> project [https://aicommons.science](https://aicommons.science)
Developed and Maintained by [John Olafenwa](https://twitter.com/johnolafenwa) and [Moses Olafenwa](https://twitter.com/OlafenwaMoses), brothers, creators of [ImageAI](https://github.com/OlafenwaMoses/ImageAI ), Authors of [Introduction to Deep Computer Vision](https://john.specpal.science/deepvision) and Co-Founders of [AICommons Global Limited](https://aicommons.science)
# Tutorials and Documentation
Visit [torchfusion.readthedocs.io](https://torchfusion.readthedocs.io) for comprehensive tutorials and examples on how to use Torchfusion
<hr>
# Installing TorchFusion
<pre> pip3 install --upgrade torchfusion </pre>
# Installing Pytorch
Visit [Pytorch.org](https://pytorch.org) for instructions on installing pytorch.
<br><br>
<h1>MNIST in Five Minutes</h1>
<pre>
from torchfusion.layers import *
from torchfusion.datasets import *
from torchfusion.metrics import *
import torch.nn as nn
import torch.cuda as cuda
from torch.optim import Adam
from torchfusion.learners import StandardLearner
#load dataset
train_loader = mnist_loader(size=28,batch_size=64)
test_loader = mnist_loader(size=28,train=False,batch_size=64)
#define model
model = nn.Sequential(
Flatten(),
Linear(784,100),
Swish(),
Linear(100,100),
Swish(),
Linear(100,100),
Swish(),
Linear(100,10)
)
#move to GPU if available
if cuda.is_available():
model = model.cuda()
#Setup optimizer and loss function
optimizer = Adam(model.parameters())
loss_fn = nn.CrossEntropyLoss()
#Define metrics
train_metrics = [Accuracy()]
test_metrics = [Accuracy()]
#Initiate Learner
learner = StandardLearner(model)
if __name__ == "__main__":
#Print summary of the model
print(learner.summary((1,28,28)))
#initiate training
learner.train(train_loader,train_metrics=train_metrics,optimizer=optimizerExcerpt of 54,660 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:47ee0256c2888c6b, topic:deep-learning, topic:neural-network, topic:pytorch