Pytorch vs Keras for building models for prediction/kaggle etc

What are people’s preferences when it comes to using models for supervised learning tasks? I get that keras, due to being a higher level library, might not be the best choice when it comes to unsupervised learning tasks like reinforcement learning or GANs (where pytorch might be better), but when it comes to supervised learning, it seems like keras has a lot of quality of life features like splitting data into training/validation during model fitting, plus a lot more documentation on things like early stopping, model checkpoint, ensemble snapshot etc.

What do you guys think is the best choice when it comes to supervised learning tasks such as kaggle competitions, is there any instance where you would want to use pytorch over keras?