plan.json

{
  "depth": 10,
  "aspect_ratio": 64,
  "head_dim": 128,
  "max_seq_len": 2048,
  "vocab_size": 16384,
  "params_total": 122552666,
  "params_scaling": 59638060,
  "flops_per_token": 515114904,
  "unique_train_tokens": 191488840,
  "target_tokens_ratio": 626199630,
  "target_tokens_epochs": 765955360,
  "num_iterations": 1194,
  "total_batch_size": 524288,
  "device_batch_size": 8,
  "world_size": 1,
  "device_batch_size_source": "ladder.json device_batch_size_by_vram",
  "grad_accum_steps": 32,
  "train_tokens": 625999872,
  "epochs": 3.27,
  "tokens_per_param": 10.5,
  "train_flops": 322461863969292288,
  "window_pattern": "L",
  "attention": "sdpa",
  "fp8": false,
  "eval_tokens": 1982464,
  "eval_every": 59,
  "save_every": 119,
  "device_type": "cuda",
  "frozen_at": "2026-10-06T14:57:22"
}