Cell-free Massive MIMO is thought to be a potential technology to meet the high demands and growing users in B5G networks and towards 6G wireless networks. The cell-free technique benefits macro-diversity since the service APs are distributed over the whole network. Power allocation is crucial that lets us to enhances cell-edge users with the price of minimizing cell center power from APs. Even while many proposed heuristics schemes as well as Deep Neural Network (DNN) accurately allocate power for minimum user rates, they are ineffective in terms of prediction time complexity as the number of users grows from small to large. In this paper, a scalable DNN-based power allocation is proposed. The input for Neural Network is done by concatenating Large-scale fading and geographical user positions The simulation result shows that the proposed DNN can successfully allocate power almost resembles with Bisection Method and achieves 400 times faster prediction time and a 34.1% improvement in execution time.