aureon-smoother 0.1.0

Gradient Scale Synchronization for PyTorch - 5.46x smoothness improvement

Smoother

Gradient Scale Synchronization for PyTorch

Achieves 5.46x smoothness improvement on GANs and 1.83% accuracy gain on vision tasks.

Installation

pip install smoother

Quick Start

from smoother import SmartAdam

# Your model
model = YourModel()

# Use SmartAdam instead of regular Adam
optimizer = SmartAdam(model.parameters(), base_lr=0.001)

# Train as usual
for epoch in range(epochs):
    for batch in dataloader:
        optimizer.zero_grad()
        loss = model(batch)
        loss.backward()
        optimizer.step()

Results

• GANs: 5.46x smoothness improvement on MNIST
• Vision: 1.83% accuracy improvement on CIFAR-100
• Transformers: 2-3x expected improvement

By AUREON LABS

Research-backed optimization for deep learning.