Thanks for write-up.

One minor thing: I believe you haven`t spent enough time on the table with ticks.

E.g., just to name a few: fast.ai HAS gradient clipping:
https://docs.fast.ai/basic_train.html#clip_grad
And multiGPU: https://twitter.com/jeremyphoward/status/1075059836440653824?lang=en
And built-in visualization: https://docs.fast.ai/vision.learner.html#_cl_int_plot_top_losses