Tune.qrandint conflict with replay buffer capacity ('Integer' and 'int')

RLlib
1

I’m attempting to tune a large set of hyperparameters, and am receiving a TyperError for the replay buffer config:
… ray\rllib\utils\replay_buffers\multi_agent_prioritized_replay_buffer.py", line 129, in __init __
shard_capacity = capacity // num_shards
TypeError: unsupported operand type(s) for //: ‘Integer’ and ‘int’

Any help is greatly appreciated

here is my script:

    config:DQNConfig = DQNConfig()
    
    explore_config = {
        "type": "EpsilonGreedy",
        "initial_epsilon": tune.uniform(0.1, 1.0),
        "final_epsilon": tune.uniform(0.01, 0.1),
        "epsilon_timesteps": tune.randint(1000, 10000),
    }
    
    
    replay_config = {
        "type": "MultiAgentPrioritizedReplayBuffer",
        "capacity": tune.qrandint(1000, 5000, 1000),
        "prioritized_replay_alpha": tune.uniform(0.2, 0.8),
        "prioritized_replay_beta": tune.uniform(0.2, 0.8),
        "prioritized_replay_eps": tune.uniform(1e-7, 1e-5), 
        "replay_sequence_length": 1,
        "worker_side_prioritization": False,
    }
    
    config=config.training(
        dueling=False,
        double_q=False,
        hiddens=tune.choice([[128, 128, 128], [256, 256, 256], [512, 512, 512]]),
        lr=tune.loguniform(1e-4, 1e-2),    
        gamma=tune.uniform(0.9, 0.99), 
        train_batch_size=tune.randint(100, 2500),
        replay_buffer_config=replay_config
    )
    
    config=config.exploration(exploration_config=explore_config)
    
    config=config.environment(env=Env, env_config=algo.setup.ENV_CONFIG)
    
    def train_dqn(config):
        model = DQNTrainer(config)
        avg_reward = model.train()
        return avg_reward          

    analysis = tune.run(
        train_dqn,
        config=config,
        metric="mean_reward",
        mode="max",
        num_samples=10000,
        scheduler=HyperBandForBOHB()
    )

    best_config = analysis.get_best_config(metric="mean_reward")
    best_checkpoint = analysis.get_best_checkpoint(metric="mean_reward")

    trainer = RLTrainer(best_config.config)
    trainer.restore(best_checkpoint)

    checkpoint_file = trainer.save()
    print("model saved to", checkpoint_file)