Setting initial iterations in Ray Tune's implementation of BOHB

Question

I am trying to use Ray Tune's implementation of BOHB to hyperparameter tune a PPO model. If I set the amount of iterations to e.g. 100 it works fine - however it already samples new hyperparameter values after only one iteration of a sample. Thereby, the bayesian search for new parameters becomes somewhat obsolete. Is there a way to define this inital search size?

My current setup is as follows:

bohb_search = TuneBOHB(
    space=hyperparams,
    metric="episode_reward_mean",
    mode="max",
    bohb_config={
        # I assume the setting goes here, but I am unable to find documentation on allowed dict keys.
    }
)
bohb_search = tune.search.ConcurrencyLimiter(bohb_search, max_concurrent=2)

bohb_hyperband = HyperBandForBOHB(
    time_attr="training_iteration",
    max_t=TRAINING_ITERATIONS,
    reduction_factor=2,
    metric="episode_reward_mean",
    mode="max",
    stop_last_trials=False,
)

tuner = tune.Tuner(
    "PPO",
    run_config=air.RunConfig(
        name="BOHB_exp_1",
        storage_path=os.path.join("~", "ray_results", "tuning"),
        stop={"training_iteration": TRAINING_ITERATIONS},
    ),
    tune_config=tune.TuneConfig(
        search_alg=bohb_search,
        scheduler=bohb_hyperband,
        num_samples=NUM_SAMPLES,
    ),
    param_space={
        "env": "biopharma_env",
        "model": {
            "custom_model": "action_mask_model",
            "vf_share_layers": True,
        },
        "framework": "tf2",
        "eager_tracing": True,
        "use_kl_loss": False,
        "num_gpus": 0,
        "num_rollout_workers": 3,
        "vf_clip_param": np.inf,
        "train_batch_size": 2048,
    },
)

I am guessing that the setting I am looking for should be specified under bohb_config in the TuneBOHB method, but I haven't been able to find the documentation for allowed keys (neither in RayTune or HpBandSter documentation).

Does anyone know how I can specify this setting?