François Chollet

@ishwors46091536 No because the share of training vs inference is roughly constant (at ~60% vs ~40%, it fluctuates a bit), so removing inference from the calculation doesn't change the fit, it's only a multiplicative constant