I am new to PyTorch and what would have been useful to me was an overview of the project, at level higher than the code.
I suggest we have a new Overview / concepts section in the documentation. That would include:
- An overview of a typical PyTorch application, main and training data, model creation, loss and optimizer , training loop and test. I know these are also standard ML concepts, but I think it would help as they tie to specific PyTorch named concepts…
- An overview of the PyTorch project in terms of python modules, models, C++ libraries, overview of the generation
- An overview of the packaging and how the C++ and python libraries sit together.
- API overviews.
- Compiler overview contrasting the newer, older and recommended approaches.
- GPU acceleration overview
- Testing overview and approaches.
- Any other overview perspective that might be useful
- A starter Glossary for main terms to be defined and linked to as the term is used
The above should contain diagrams of the concepts and also link down to the existing doc. I hope that such an overview would make it more approachable to beginners and paint the big picture.
Please feedback if you think this would be useful. I am happy to put up initial content for it to be reviewed.