Intelligent reflecting surface (IRS) has been acknowledged as a pivotal enabling technology for the sixth-generation (6G) wireless communication systems as it can manipulate wireless propagation environment via controllable signal reflection. Given the intricacy of the communication environment and the non-convex nature of hybrid beamforming sum rate maximization, there are challenges in the processing of traditional methods. By interacting with the environment, deep reinforcement learning (DRL) can effectively acquire the optimal beamforming strategy. Its end-to-end learning style, strong adaptability and real-time performance make it ideal for dealing with the complexity and real-time requirements of communication systems. Using deep reinforcement learning can improve system performance and enhance its adaptability. DRL merges the powerful representation learning capabilities of deep neural networks (DNN) with the decision optimization abilities of reinforcement learning (RL). This amalgamation empowers computer systems to acquire decision-making skills through interaction with the environment, with the aim of maximizing a certain reward signal. In this paper, by learning the knowledge of DRL, combining Long short-term memory (LSTM) and deep deterministic policy gradient (DDPG) algorithm, the optimization problem jointly designed by transmit beamforming matrix at the base station(BS) and the phase shift matrix at the IRS is studied for multiple input single output (MISO) systems. The simulation outcomes underscore the robustness of the LSTM-DDPG algorithm in both maximizing the achievable sum rate and runtime efficiency.