training.json

{
 "run": "t24h-adamw-int8",
 "step": 10303,
 "games": 430122,
 "positions": 51834880,
 "seconds": {
  "self_play": 74282.77617749186,
  "gradient": 11168.833350930203
 },
 "train_config": {
  "n_sims": 128,
  "batch_games": 1024,
  "e_cap": 96,
  "tau_plies": 30,
  "eps": 0.25,
  "alpha": 0.3,
  "terminal_collapse": true,
  "gumbel": true,
  "gumbel_m": 16,
  "gumbel_scale": 1.0,
  "fp8": false,
  "int8": true,
  "pcr_p": 0.0,
  "pcr_fast_sims": 64,
  "sims_schedule": [],
  "window_games": 20000,
  "mean_plies": 350,
  "max_plies": 512,
  "buffer_dir": null,
  "buffer_in_memory": false,
  "samples_per_position": 0.8150000000000001,
  "samples_per_game": 65.2,
  "batch": 4096,
  "micro_batch": 1024,
  "momentum": 0.9,
  "l2": 0.0001,
  "optimizer": "adamw",
  "betas": [
   0.9,
   0.95
  ],
  "adam_wd": 0.01,
  "aux_wd": null,
  "grad_clip": 1.0,
  "aux_lr": 0.001,
  "muon_momentum": 0.95,
  "muon_qkv_split": true,
  "muon_head_group": 8,
  "muon_ns_steps": 5,
  "muon_ns_scheme": "polar",
  "normuon": false,
  "lr_schedule": [
   [
    0.0,
    0.001
   ]
  ],
  "total_steps": 9800,
  "warmup_steps": 30,
  "decay": "cosine",
  "lr_min": 5e-05,
  "moves_per_phase": 10,
  "checkpoint_every": 200,
  "keep_checkpoints": true,
  "puzzle_probe": true,
  "puzzle_limit": 20000,
  "buffer_snapshot_every": 10000,
  "autocast": true,
  "value_weight": 1.0,
  "compile": false,
  "min_records": 0,
  "euros_per_hour": 0.0,
  "strict_labels": true,
  "audit_every": 100,
  "collect_search_stats": true,
  "deterministic": true,
  "seed": 0,
  "impl": "cuda",
  "encoder": "cuda"
 }
}