training_meta.json

{
  "step": 10358,
  "val_loss": 1.8148479143190384,
  "val_ppl": 6.140142278710768,
  "params_M": 49.443,
  "pushed_at": "2026-07-25T11:18:48.061114",
  "tokens_seen": 12206861568,
  "architecture": "Rose X1",
  "features": [
    "GQA",
    "QK-Norm",
    "RoPE",
    "SwiGLU",
    "RMSNorm",
    "XSA-RefreshGate"
  ],
  "optimizer": "muon_adamw"
}