The paper describes an online deep learning algorithm for the adaptive modulation and coding in 5G Massive MIMO. The algorithm is based on a fully-connected neural network, which is initially trained on the output of the traditional algorithm and then is incrementally retrained by the service feedback of its own output. We show advantage of our solution over the state-of-the-art Q-Learning approach. We provide system-level simulation results to support this conclusion in various scenarios with different channel characteristics and different user speed. Compared with traditional OLLA the proposal shows 10% to 20% improvement of user throughput in full buffer case.