It's ok but not great. Like say your model doesn't always converge, that would be one way to deal with it.

I'd prefer to see someone tune hyperparameters so that the metrics are minimally sensitive to the random seed though