I really like the standardization of logging etc. that FluxTraining comes with. However, for my applications, the logging is much too verbose. Logging the loss after every gradient step significantly slows down my training if the network is fairly small.
In those settings, it would be helpful if the default-callbacks could be told to run at a lower frequency; e.g. every N training examples or even only on at the end of each epoch. Is something like that currently supported without implementing a custom callback?
I really like the standardization of logging etc. that FluxTraining comes with. However, for my applications, the logging is much too verbose. Logging the loss after every gradient step significantly slows down my training if the network is fairly small.
In those settings, it would be helpful if the default-callbacks could be told to run at a lower frequency; e.g. every N training examples or even only on at the end of each epoch. Is something like that currently supported without implementing a custom callback?