Chapter 8. PyTorch to Production
In 2017, when PyTorch released its usable version, the promise was for it to be a Python-first framework for researchers. The PyTorch community was strict about this for a year, but then it saw the abundance of production requirements and decided to merge production capability with PyTorch's first stable release, 1.0, but without compromising the usability and flexibility it was created for.
PyTorch is known for being a clean framework, and hence it was a challenging task to achieve the production capability and flexibility needed for research. I think that the major hurdle for pushing production support to the core was going out of Python's realm and moving the PyTorch model to a faster, thread-safe language that has multithreading capability. But then, that violated the Python-first principle that PyTorch had up to that point.
The first step toward solving this problem was to make the Open Neural...
