{
  "model_type": "simclr-resnet18",
  "framework": {
    "training": "Lightly",
    "backend": "PyTorch Lightning"
  },
  "architecture": {
    "backbone": "resnet18",
    "backbone_feature_dim": 512,
    "projection_head": {
      "implementation": "lightly.models.modules.heads.SimCLRProjectionHead",
      "input_dim": 512,
      "hidden_dim": 512,
      "output_dim": 128
    }
  },
  "objective": {
    "loss": "NTXentLoss",
    "temperature": 0.1
  },
  "optimizer": {
    "name": "LARS",
    "momentum": 0.9,
    "weight_decay": 1e-06
  },
  "scheduler": {
    "name": "CosineWarmupScheduler",
    "interval": "step",
    "warmup_epochs": 10
  },
  "input": {
    "size": [
      224,
      224
    ],
    "channels": 3
  },
  "paper": {
    "title": "Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field",
    "authors": [
      "Dylan M. Diaz",
      "Margaret M. Henderson"
    ],
    "year": 2026,
    "doi": "10.32470/0416gfsq",
    "arxiv": "2607.19316"
  },
  "notes": [
    "This configuration describes the reference-model training setup used in the associated study.",
    "The exact installed Lightly and PyTorch Lightning package versions were not stored in this configuration."
  ],
  "repository": "DM-Diaz/SimCLR-ResNet18-ImageNet100",
  "checkpoint": "checkpoint_120-resnet18-simclr-imagenet100.ckpt",
  "training": {
    "dataset": {
      "huggingface_id": "clane9/imagenet-100",
      "num_classes": 100,
      "local_format": "ImageFolder-style train/val directories"
    },
    "epochs": 120,
    "batch_size": 64,
    "base_learning_rate": 0.3,
    "lr_scaling_rule": "base_learning_rate * batch_size / 256",
    "initial_learning_rate": 0.075,
    "distributed": false,
    "mixed_precision": "16-mixed",
    "knn_k": 20,
    "knn_temperature": 0.1
  }
}