cartpole-appo-vtrace-fake-gpus:
    env: CartPole-v0
    run: APPO
    stop:
        episode_reward_mean: 150
        training_iteration: 400
    config:
        # Works for both torch and tf.
        framework: tf
        num_envs_per_worker: 5
        num_workers: 1
        observation_filter: MeanStdFilter
        num_sgd_iter: 6
        vf_loss_coeff: 0.01
        vtrace: true
        vtrace_drop_last_ts: false

        # Double batch size (2 GPUs).
        train_batch_size: 1000
        lr: 0.0003

        model:
            fcnet_hiddens: [32]
            fcnet_activation: linear
            vf_share_layers: true
            # Test w/ LSTMs.
            use_lstm: true

        # Fake 2 GPUs.
        num_gpus: 2
        _fake_gpus: true