training_meta.json

{
  "step": 19075,
  "val_loss": 1.6259081413888932,
  "val_ppl": 5.083033054470817,
  "params_M": 151.274,
  "pushed_at": "2026-08-16T19:41:01.107398",
  "tokens_seen": 79967283200,
  "architecture": "Rose X1",
  "features": [
    "GQA",
    "QK-Norm",
    "RoPE",
    "SwiGLU",
    "RMSNorm",
    "XSA-RefreshGate"
  ],
  "optimizer": "muon_adamw"
}