models/checkpoints/smoke-train/run_config.json

{
  "data": "training/data/tinyshakespeare.txt",
  "out_dir": "models/checkpoints",
  "run": "smoke-train",
  "reset": true,
  "steps": 1,
  "batch": 1,
  "eval_batch": 0,
  "accum": 1,
  "ctx": 4,
  "eval_steps": 1,
  "eval_interval": 1,
  "save_interval": 0,
  "no_save": true,
  "seed": 42,
  "lr": 0.0001,
  "weight_decay": 0.0,
  "grad_clip": 1.0,
  "optimizer": "signsgd",
  "tscale_type": "T32",
  "compute_dtype": "none",
  "max_moe_iters": 1,
  "accum_threshold": 3,
  "freeze_scales": false,
  "ternary_updates": true,
  "allow_float_trainables": false,
  "enable_image": false,
  "enable_audio": false,
  "disable_vq": true,
  "disable_graph": true,
  "disable_memory": true,
  "disable_moe": false,
  "cpu": false,
  "audit": "Ternary state audit:\n  logical ternary weights: 1,246,263,744\n  ternary training state: 1625.18 MB (T=237.71, E=99.48, E_accum=99.46, accum=1188.53)\n  trainable float params: 0 tensors, 0.00 MB\n  frozen float params: 0 tensors, 0.00 MB\n  float buffers: 0 tensors, 0.00 MB"
}