plan.json

{
  "depth": 8,
  "aspect_ratio": 64,
  "head_dim": 64,
  "max_seq_len": 1024,
  "vocab_size": 16384,
  "params_total": 75497898,
  "params_scaling": 33554816,
  "flops_per_token": 251660688,
  "unique_train_tokens": 75776124,
  "target_tokens_ratio": 352325568,
  "target_tokens_epochs": 454656744,
  "num_iterations": 1344,
  "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": 352321536,
  "epochs": 4.65,
  "tokens_per_param": 10.5,
  "train_flops": 88665480146976768,
  "window_pattern": "L",
  "attention": "sdpa",
  "fp8": false,
  "eval_tokens": 786432,
  "eval_every": 67,
  "save_every": 134,
  "device_type": "cuda",
  "frozen_at": "2026-10-05T13:18:30"
}