qixuanai/torchservant

Many useful tools for PyTorch.

TorchServant

中文版本 | English Version

TorchServant is still in design and development stage, and more modules may be added. Welcome interested friends to join this project.````

TorchServant is an assembly helping quickly development for PyTorch users. The essential design idea of torchservant is to liberate researchers from tedious repetitions such as save&load checkpoints, make records, manage GPUs, etc. It could help users focus on the core pipelines of design neural networks, including design models, choose hyper-parameters, design loss functions and optimizers, train models, evaluate models, and etc.

Packages

TorchServant contains the following components:

Package Name	Explaination
cfgenator	Configure file generator
modelservant	Manage weights files and checkpoints, keep training process continuous.
procmonitor	API for visdom and tensorboard, visualize diagrams, illustrations and progresses.
stats	Resource statistics on GPUs, memories, cpu, and consumed time.
classicmodels	Include AlexNet, VGG, Resnet, Inception, etc.
visualfeature	Visualize feature maps during training and evaluation process.
weightransfer	An Qt-based visual tool to transfer weights between different models.

Install

pip install torchservant

Example