Aurane
A modern, elegant ML-oriented DSL that transpiles to idiomatic Python
Powerful Features
Everything you need to build ML models faster
Elegant Syntax
Write expressive, clean .aur files that compile into production-ready PyTorch code
Lightning Fast
Automatic recompilation with watch mode and sub-100ms compilation times
Model Inspection
Detailed architecture analysis with parameter counts and shape inference
Interactive REPL
Live coding environment for rapid experimentation and prototyping
Auto-Format
Built-in formatting and linting tools to maintain code quality
Complete Pipeline
Define experiments, datasets, models, and training in one unified DSL
See It In Action
Write elegant DSL, get production-ready code
Simple Model
A basic neural network in Aurane
model SimpleNet:
input_shape = (1, 28, 28)
def forward(x):
x -> flatten()
-> dense(128).relu
-> dense(10)Training Pipeline
Complete experiment configuration
experiment MnistBaseline:
seed = 42
device = "auto"
train MnistCNN on mnist_train:
validate_on = mnist_test
loss = cross_entropy
optimizer = adam(lr=1e-3)
epochs = 5Generated PyTorch
Clean, idiomatic output
class SimpleNet(nn.Module):
def __init__(self):
super().__init__()
self.dense1 = nn.Linear(784, 128)
self.dense2 = nn.Linear(128, 10)
def forward(self, x):
x = torch.flatten(x, 1)
x = F.relu(self.dense1(x))
x = self.dense2(x)
return xGet Started in Seconds
Install Aurane and start building ML models
pip install -e ".[all]"Clone repository
$ git clone https://github.com/desenyon/aurane.gitNavigate to directory
$ cd auraneInstall with all features
$ pip install -e ".[all]"Verify installation
$ aurane --helpRequirements
- •Python 3.10 or higher
- •PyTorch 2.0+ (for ML features)
- •Rich 13.0+ (for beautiful CLI)