ARS always stops after 2,464M num of steps, despite exponential reward grow

if __name__ == "__main__":

    env = CustomEnv()
    #check_env(env)
     # Simplified architecture
    policy_kwargs = dict(
        net_arch=[2048,1024, 512, 256,128,2], 
        activation_fn=torch.nn.GELU,
    )

    model = ARS(ARSPolicy, env, verbose=0, tensorboard_log=logdir, device='cuda',
                policy_kwargs=policy_kwargs,stats_window_size =1000,delta_std=0.07,learning_rate=0.0035,n_eval_episodes=500)

    TIMESTEPS = 12288000  
    #vec_env = model.get_env()
    observation,info = env.reset()
    done = False

    while True:

        model.learn(total_timesteps=TIMESTEPS, reset_num_timesteps=False, tb_log_name=f"ARS")

        model.save(f"{models_dir}/model_{int(time.time())}")

        action, _ = model.predict(observation)

        observation, reward, terminated, truncated, info = env.step(action)

I tried to find some limit/early stopping or smth in the stablabaselines contrib library, but I didn’t find anything