{
  "experiment": "clip8_vit",
  "task": "MNIST",
  "seed": 42,
  "steps": 4000,
  "epochs": null,
  "train": {
    "optimizer": "adam",
    "method": "adam",
    "batch_size": 128,
    "learning_rate": 1e-05,
    "weight_decay": 0.0,
    "max_steps": 4000,
    "schedule": "cosine",
    "warmup_steps": 0,
    "num_workers": 8,
    "fused": true,
    "gradient_checkpointing": false,
    "log_every": 20
  }
}