v0.4.0

What's Changed

Please refer to CHANGELOG.md for details.

New Contributors

• @afzal442 made their first contribution in #179
• @renovate made their first contribution in #182
• @rwiteshbera made their first contribution in #184
• @sayantan1410 made their first contribution in #194
• @theory-in-progress made their first contribution in #235
• @dependabot made their first contribution in #255
• @guptaaryan16 made their first contribution in #265
• @puhuk made their first contribution in #321

Thanks to our maintainers (in alphabetical order):
@guptaaryan16 @theory-in-progress @trsvchn @vfdev-5 @ydcjeff

Full Changelog: v0.2.0...v0.4.0

v0.2.0

Please refer to CHANGELOG.md for details.

v0.1.0

Today we are pleased to announce the official release of Code-Generator v0.1.0. Code-Generator is a web application which can bootstrap quick-start Python code for common training tasks in deep learning. It is powered by PyTorch and PyTorch-Ignite with full of best practices.

Features

The codes are generated through the template files which are

• Distributed configurations ready (Multi-GPU, Multi-Node Single/Multi-GPU Training)
• Device agnostic (CPU, GPU, TPU)
• Full of features provided by PyTorch and PyTorch-Ignite (Checkpointing, Resuming, Ignite Handlers, Experiment Tracking Systems)

The following templates are available as part of v0.1.0 release:

1. Image Classification
2. Text Classification
3. Generative Adversarial Network
4. Single Model, Single Optimizer

Code-Generator is not an another abstraction over PyTorch and PyTorch-Ignite. It is a web app which generates the code which means you still have the full control over the generated code. You can start with the minimal template and enable or disable options according to your needs.

Trying It Out

1. Go to Code-Generator web app.
2. Configure the available options.
3. Grab the codes.
4. Solve the problems.