relayrl_framework 0.5.0-rc.1

A heterogeneous RL runtime control platform for concurrent multi-actor execution.
Documentation
{
    "client_config": {
        "config_polling_seconds": 10,
        "init_hyperparameters": {
            "PPO": {
                "discrete": true,
                "seed": 0,
                "traj_per_epoch": 1,
                "clip_ratio": 0.1,
                "gamma": 0.99,
                "lam": 0.97,
                "pi_lr": 3e-4,
                "vf_lr": 3e-4,
                "train_pi_iters": 40,
                "train_v_iters": 40,
                "target_kl": 0.01
            },
            "IPPO": {
                "discrete": true,
                "seed": 0,
                "traj_per_epoch": 1,
                "clip_ratio": 0.1,
                "gamma": 0.99,
                "lam": 0.97,
                "pi_lr": 3e-4,
                "vf_lr": 3e-4,
                "train_pi_iters": 40,
                "train_v_iters": 40,
                "target_kl": 0.01
            },
            "MAPPO": {
                "discrete": true,
                "gamma": 0.99,
                "lam": 0.97,
                "clip_ratio": 0.2,
                "pi_lr": 3e-4,
                "vf_lr": 1e-3,
                "train_pi_iters": 80,
                "train_vf_iters": 80,
                "target_kl": 0.01,
                "traj_per_epoch": 8
            },
            "CUSTOM": {
                "_comment": "Add custom algorithm hyperparams here formatted just like the other algorithms. i.e. \"MAPPO\": {...}",
                "_comment2": "Make sure to add the algorithm name to the algorithm_name field",
                "_comment3": "These key-values will be sent to the server for initialization"
            }
        },
        "local_model_module": {
            "directory": "model_module",
            "model_name": "client_model",
            "format": "pt"
        },
        "metrics": {
            "meter_name": "relayrl-client",
            "otlp_endpoint": {
                "prefix": "http://",
                "host": "127.0.0.1",
                "port": "4317"
            }
        },
        "trajectory_file_output": {
            "directory": "experiment_data",
            "_comment": "use `Csv` or `Arrow`",
            "file_type": "Csv"
        }
    },
    "transport_config": {
        "nats_addresses": {
            "inference_server_address": {
                "host": "127.0.0.1",
                "port": "50050"
            },
            "training_server_address": {
                "host": "127.0.0.1",
                "port": "50051"
            }
        },
        "zmq_addresses": {
            "inference_addresses": {
                "inference_server_address": {
                    "host": "127.0.0.1",
                    "port": "7800"
                },
                "inference_scaling_server_address": {
                    "host": "127.0.0.1",
                    "port": "7801"
                }
            },
            "training_addresses": {
                "model_server_address": {
                    "host": "127.0.0.1",
                    "port": "50051"
                },
                "trajectory_server_address": {
                    "host": "127.0.0.1",
                    "port": "7776"
                },
                "agent_listener_address": {
                    "host": "127.0.0.1",
                    "port": "7777"
                },
                "training_scaling_server_address": {
                    "host": "127.0.0.1",
                    "port": "7778"
                }
            }
        }
    }
}