I work on a neural network library in C++ and I'm in the process of implementing recurrent neural networks. I run my tests on a GitHub server on different datasets, Iris and Wine for simple unit tests, MNIST and Fashion MNIST for test the MLP, CIFAR-10 for test the CNN and I looking for some new dataset to train the RNN that I implement.
Do you have any idea of small datasets to train my RNN?
I'm looking for one or two small datasets that are easy to integrate into my tests in my continuous delivery process. About 30 seconds for 1 epoch with more than 70-80% accuracy would be ideal.
PS: I know I can use MNIST to simulate temporal data but having a "real" dataset would be better and more diversified.