{
    "checkpoint_freq": 1,
    "dataloader_num_workers": 0,
    "gradient_accumulation_steps": 1,
    "keep_checkpoints": -1,
    "learning_rate": 0.0001,
    "logging_dir": "logs",
    "mixed_precision": "bf16",
    "model": {
        "type": "vlmpar",
        "model_name": "google/siglip2-base-patch16-224",
        "num_classes_dict": {
            "upper_body_color": 11,
            "lower_body_color": 11,
            "gender": 2,
            "bag": 2,
            "hat": 2
        }
    },
    "num_train_epochs": 10000,
    "optimizer": {
        "type": "AdamW",
        "lr": 0.0001,
        "weight_decay": 0.01
    },
    "project_dir": "outputs",
    "scheduler": {
        "type": "LinearLR",
        "start_factor": 1.0,
        "end_factor": 0.0,
        "total_iters": 10000
    },
    "train_batch_size": 32,
    "train_dataset": {
        "type": "mivia_par_kd_2025",
        "root_dir": "/mnt/dati/homes/zsellam/datasets/par2025",
        "split": "training",
        "processor": "google/siglip2-base-patch16-224"
    },
    "val_dataset": {
        "type": "mivia_par_kd_2025",
        "root_dir": "/mnt/dati/homes/zsellam/datasets/par2025",
        "split": "validation",
        "processor": "google/siglip2-base-patch16-224"
    },
    "valid_batch_size": 32,
    "warmup_ratio": 0.1,
    "weight_decay": 0.01,
    "config": "./configs/siglip_par_train_config.py",
    "seed": null,
    "dtype": null,
    "report_to": "tensorboard",
    "max_grad_norm": 1.0,
    "max_train_steps": null,
    "validation_split": 0.25
}