TorchFusion-Utils Alpha Release

Alpha release of TorchFusion Utils

Whats New!

• Mixed Precision Training
  Train your pytorch models faster with reduced memory usage. With few lines of code, you can take advantage of nvidia tensor cores for accelerated training of deep learning models.

Example

from torchfusion_utils.fp16 import convertToFP16

#convert your model and optimizer to mixed precision mode
model, optim = convertToFP16(model,optim)

#in your batch loop, replace loss.backward with optim.backward(loss)
optim.backward(loss)

• Initializers
  Easy initialize your model with state-of-the-art initializers with fine grained control over layers and parameters.

Example

from torchfusion_utils.initializers import *

#initialize the convolution layers with kaiming_normal
kaiming_normal_init(model,types=[nn.Conv2d])

#initialize the linear layers with normal
normal_init(model,types=[nn.Linear])

#initialize batchnorm weights with ones
ones_init(model,types=[nn.BatchNorm2d],category="weight")

#initialize batchnorm bias with zeros
zeros_init(model,types=[nn.BatchNorm2d],category="bias")
 

• Metrics
  Support for popular metrics and simple api to create your custom metrics

• Model Utils
  Functions to reliably load and save models and summary function for analyzing the parameters, computational cost and structure of deep learning models

Documentation

Comprehensive documentation from utils.torchfusion.org