resnet18_cifar10_accuracy_aware.json

{
    "model": "resnet18",
    "weights": "https://storage.openvinotoolkit.org/repositories/nncf/examples/torch/classification/resnet18_cifar10.pth",
    "batch_size": 512,
    "input_info": {
        "sample_size": [
            1,
            3,
            32,
            32
        ]
    },
    "dataset": "cifar10",
    "num_classes": 10,
    "optimizer": {
        "type": "SGD",
        "base_lr": 0.001,
        "weight_decay": 1e-4,
        "schedule_type": "multistep",
        "steps": [
            60,
            100
        ],
        "optimizer_params": {
            "momentum": 0.9,
            "nesterov": true
        }
    },
    "accuracy_aware_training": {
        "mode": "adaptive_compression_level",
        "params": {
            "maximal_relative_accuracy_degradation": 1.0,
            "initial_training_phase_epochs": 110,
            "patience_epochs": 110,
            "maximal_total_epochs": 1000
        }
    },
    "compression": {
        "algorithm": "filter_pruning",
        "pruning_init": 0.1,
        "params": {
            "schedule": "exponential",
            "pruning_target": 0.4,
            "pruning_steps": 20,
            "filter_importance": "geometric_median"
        }
    }
}