After a custom model is built using Torch and PL, are there any easy methods to test this? Eg. if there’s no bug in the architecture
For example, lets say i build a custom RNN for sequence classification with PL on my dataset. Now, lets say I see poor performace. How do I assure myself that there is no bug in the model? If my model is general enough, I would like to run in with any of the built datasets that flash contains that most matches. Based on the performance on that relatively easy dataset, we could conclude that the model has passed sanity check. Would this also be a good feature?