train_log.json

{
  "king_dir": "/root/king_165_hf",
  "source_dirs": [
    "/root/datasets/finewebedu"
  ],
  "source_weights": [
    1.0
  ],
  "lora_r": 128,
  "lr": 5e-05,
  "muon_lr": 0.0001,
  "optimizer": "muon_adamw",
  "steps": 5000,
  "unfreeze_head": false,
  "grad_accum": 4,
  "seq_len": 2048,
  "initial_loss": 2.1343601644039154,
  "final_loss": 2.006874531507492,
  "min_loss": 1.2580181956291199,
  "min_eval_ce": 1.9948618136644363,
  "best_eval_count": 9,
  "out_dir": "/root/output_muonadamw_r128_fwe_muonlr1e-4_k165/final_merged"
}