Training frequency in DQN rllib

I want to specify the training frequency based on timesteps for DQN in RLlib but I couldn’t find the corresponding parameters. I run my code for some iterations (using algorithm.train()) and in each iterations I have 1000 timesteps (episode_size=1000, so one episode in each iteration), the train_batch_size=32, but I don’t know how many times (every how many timesteps) the network is updated in each iteration. I am confused with these parameters, any helps?