Add time and cost to your tutorials
It would seem that GCP enables any user to train a resnet50 in 30 minutes for 12.60$. However it is only if ones follows this very specific tutorial: https://cloud.google.com/blog/products/ai-machine-learning/how-to-train-a-resnet-image-classifier-from-scratch-on-tpus-on-cloud-ml-engine , from my point of view this tutorial allows no easy modification to the code.
However, this tutorial https://cloud.google.com/tpu/docs/tutorials/resnet would seem to be more straightforward if one wished to modify the source code. It also uses TPU's and one would expect it to be similar in time and money to the first. But no, following exactly that tutorial leads to a training of more than 4 days and up to 250$ cost. Which is like being stabbed in the back compared to 12.60$ (thanks for that, I could have stoped the TPU cloud to save money, but it is better to know and not regret not doing it).
So a bit of head's up in every one of your tutorials would be very much appreciated. (because there are about 3/4 tutorials to train resnet50)