A general question on approaching /learning PyTorch.
I started learning PyTorch something like 6 months ago. First I started by writing my own scripts based on some examples. However, I pretty fast understood that there is a large number of ways to structure code and a pretty large number of different tricks for training. There is PyTorch Lightning which tries to simplify training, there is fast.ai, there are training scripts that are provided by PyTorch developers themself, there are training scripts from timm.
I would like to hear the opinions of the people who have more experience. Are there some training scripts that are multitasking suitable vision/NLP for standard tasks and/or which are commonly used? I do not want to reinvent the wheel by myself, but would rather use some already field-proven tools. I find timm scrips pretty complicated for me at this moment. fast.ai - easy to use, but hide a lot of important features behind the simple inferface.