plan.json

{
  "depth": 6,
  "aspect_ratio": 64,
  "head_dim": 64,
  "max_seq_len": 1024,
  "vocab_size": 16384,
  "params_total": 42074366,
  "params_scaling": 16908504,
  "flops_per_token": 129762720,
  "unique_train_tokens": 20758910,
  "target_tokens_ratio": 177539292,
  "target_tokens_epochs": 83035640,
  "num_iterations": 316,
  "total_batch_size": 262144,
  "device_batch_size": 16,
  "world_size": 1,
  "device_batch_size_source": "ladder.json device_batch_size_by_vram",
  "grad_accum_steps": 16,
  "train_tokens": 82837504,
  "epochs": 3.99,
  "tokens_per_param": 4.9,
  "train_flops": 10749219837050880,
  "window_pattern": "L",
  "attention": "sdpa",
  "fp8": false,
  "eval_tokens": 147456,
  "eval_every": 50,
  "save_every": 100,
  "device_type": "cuda",
  "frozen_at": "2026-10-04T11:56:42"
}