{
  "config": {
    "num_workers": 16,
    "model": {
      "checkpoint_dir": ".cache/nanochat/base_checkpoints",
      "model_tag": "d12",
      "step": 1680
    },
    "data": {
      "train_path": "data/lm/climbmix_train_d12.jsonl",
      "seed": 42
    },
    "subsets": {
      "strategy": "expander",
      "n_subsets": 1000,
      "d": 10
    },
    "stage1": {
      "layer_idx": 8,
      "rank": 32,
      "lr": 0.0003,
      "lr_warmup_steps": 100,
      "lr_final_ratio": 0.1,
      "iters": 10000,
      "log_every": 25,
      "subsets_per_iter": 8,
      "batch_size_fid": 2,
      "batch_size_stab": 2,
      "fidelity_mode": "all_tokens",
      "answer_weight": 5.0,
      "stability_mode": "truncated_kl",
      "stability_top_m": 20,
      "lambda_fid": 1.0,
      "lambda_stab": 1.0,
      "lambda_lds": 0.1,
      "sketch_dim": 64,
      "muon_lr": 0.0003,
      "a_in_muon": true
    },
    "stage2": {
      "eval_span": "full",
      "lasso_alpha": 0.01
    },
    "run_name": "d12_mlr3e4",
    "output_dir": "outputs/lm/sweep_d12_mlr3e4"
  },
  "n_train": 1317003,
  "n_subsets": 1000,
  "elapsed_s": 2303.782348871231,
  "operator": {
    "d_model": 768,
    "rank": 32,
    "n_subsets": 1000
  }
}