training_metadata.json

{
  "checkpoint_id": "ckpt_20260425_135103_18000_22deff1b_9470fbb7",
  "created_at": "2026-04-25T13:51:04.065681",
  "iteration": 18000,
  "epoch": 0,
  "train_loss": 0.0,
  "val_loss": 0.00032569289276580094,
  "learning_rate": 0.00011444514640904435,
  "model_config": {
    "n_layer": 4,
    "n_head": 4,
    "n_embd": 256,
    "vocab_size": 50257,
    "block_size": 1024,
    "dropout": 0.1,
    "bias": true,
    "initial_connections": 0.1,
    "connection_growth_rate": 0.05,
    "max_connections": 1.0
  },
  "training_config": {
    "learning_rate": 0.0002,
    "batch_size": 2,
    "max_iters": 500,
    "warmup_iters": 5000,
    "lr_decay_iters": 50000,
    "min_lr": 1e-05,
    "weight_decay": 0.1,
    "grad_clip": 1.0,
    "enable_curriculum_learning": true,
    "enable_introspection": true
  },
  "data_config": {
    "data_dir": "data/nanecho",
    "batch_size": 2,
    "block_size": 1024
  },
  "metrics": {
    "val_loss": 0.00032569289276580094,
    "connection_ratio": 1.0,
    "tokens_processed": 36864000,
    "training_speed_iters_per_sec": 0.08497924105645045
  },
  "tags": [
    "phase_adaptive_mastery",
    "high_quality",
    "nanecho",
    "curriculum",
    "introspection"
  ],
  "parent_checkpoint": null,
  "notes": "Training checkpoint at iteration 18000 (resumed from iteration 17500) | Phase: adaptive_mastery",
  "file_size_mb": 253.3669786453247,
  "quality_score": 2764800.7811699593
}