hyperparams.json

{
  "baseline": {
    "learning_rate": 0.0003,
    "batch_size": 32
  },
  "large-lr": {
    "learning_rate": 0.001,
    "batch_size": 32
  },
  "cosine-sched": {
    "learning_rate": 0.0005,
    "batch_size": 64
  },
  "warmup-long": {
    "learning_rate": 0.0005,
    "batch_size": 128
  },
  "dropout-0.2": {
    "learning_rate": 0.0003,
    "batch_size": 64
  }
}