{
  % This key enables more verbose logging in the manager and project observer.
  "verbose": true, % default=false, optional
  % This ID should be the same as for the orchestrator.
  "customerID": "Admin", 
  % FLOps has only been tested for GitHub so far.
  "ml_repo_url": "https://github.com/Malyuk-A/flops_ml_repo_mnist_sklearn",
  % Supported flavors include: sklearn, pytorch, tensorflow, keras.
  "ml_model_flavor": "sklearn",
  % This key only works for special repositories intended for development.
  % It tells the builder to use prebuilt base images to significantly speed up image builds and development.
  "use_devel_base_images": false, % default=false, optional
  % This key expects a list of target platforms on which the built images should run.
  % It supports linux/amd64 and linux/arm64.
  "supported_platforms": ["linux/amd64"], % default=["linux/amd64"], optional
  "training_configuration": {
      % This key specifies the FL type.
      % FLOps supports classic and hierarchical modes.
      "mode": "classic", % default="classic", optional
      % Requested data tags should match available data on learner nodes.
      % Multiple different tags can be requested.
      % The ML data server will use local data fragments that match any of the provided tags.
      % If no data tags are provided the learner will notify watchers that it cannot find any data.
      "data_tags": ["mnist"],
      % Training cycles only apply to the hierarchical mode.
      "training_cycles: 1, % default=1, optional
      % This key tells learners the number of training and evaluation rounds to perform.
      "training_rounds": 3, % default=3, optional
      % Clients mean learners in this context.
      "min_available_clients": 2, % default=1, optional
      "min_fit_clients": 2, % default=1, optional
      "min_evaluate_clients": 2 % default=1, optional
  },
  % FLOps supports these two post-training steps.
  "post_training_steps": ["build_image_for_trained_model", "deploy_trained_model_image"], % default=[], optional
  % These are optional values that require fine-tuning.
  % Note that these values are not recommendations but placeholder values.
  "resource_constraints": {
      "memory": 250, % in MB
      "vcpus": 1,
      "storage": 25 % in MB
  }
}

{
  "verbose": true,
  "customerID": "Admin",
  "ml_repo_url": "https://github.com/Malyuk-A/flops_ml_repo_mnist_sklearn",
  "ml_model_flavor": "sklearn",
  "training_configuration": {
      "data_tags": ["mnist"],
      "min_available_clients": 2,
      "min_fit_clients": 2,
      "min_evaluate_clients": 2
  },
  "post_training_steps": ["build_image_for_trained_model", "deploy_trained_model_image"],
}

{
  "customerID": "Admin",
  "ml_repo_url": "https://github.com/Malyuk-A/flops_ml_repo_cifar10_pytorch",
  "ml_model_flavor": "pytorch",
  "supported_platforms": ["linux/amd64", "linux/arm64"],
  "training_configuration": {
      "mode": "hierarchical",
      "data_tags": ["cifar10"],
      "training_cycles": 10,
      "training_rounds": 5,
      "min_available_clients":3,
      "min_fit_clients": 3,
      "min_evaluate_clients": 3
  },
  "post_training_steps": ["build_image_for_trained_model", "deploy_trained_model_image"],
}