PyTorch Issues

There are a myriad of issues in using torch for real world applications. It is easily understood by recalling that the primary goal and purpose of torch was for research/learning and prototyping whereas the main focus of tflow has been engineering and commercial applications. Most notably, torch chose an OO or class approach which is not the best approach for ML since it quickly leads to applications with several orders of magnitude higher LOC which severely impacts performance but more importantly maintainability. Ease of learning is often cited as an advantage of torch but in practice tflow is much easier to learn/use and Keras has excellent documentation with a good library of code examples. Even if you ignore these concerns with torch you then have to address deployment issues for mobile, IoT, and edge devices which are a staple for AI applications. Thus, torch just does not have the reach or scalability of tflow. There are issues with both torch and tflow. However, these are the core issues with torch that I doubt will ever be fixed/addressed.