Keras Tuner and max model Size

Hello mentors and readers,

I am trying to restraint my model size to a certain max number of parameter.
For exemple, let’s say I have a Neural net with 5 inputs, and max number of parameter 256.
I tried the argument max_model_size in a tuner , but I feel it is not working at all, as the tuner returns too big models for my application.
And of course, google the issue leads to an open GitHub thread :

Can someone help me with restraining keras tuner to a certain model size ?

Set keras_tuner.engine.tuner.MAX_FAIL_STREAK to a really high number to increase the number of before calling This way, the tuner will continue to look for an architecture that satisfies your constraints.

One way to determine MAX_FAIL_STREAK is to use the number of combinations of hyperparameters the tuner has to evaluate to pick the best model.