Rose-1.5-Medium / training_meta.json
GODELEV's picture
Duplicate from GODELEV/Rose-1.5-Mini
5cb0481 verified
Raw History Blame Contribute Delete
361 Bytes
{
"step": 153000,
"val_loss": 2.6666615019321442,
"val_ppl": 5.294462705905101,
"params_M": 98.612,
"pushed_at": "2026-09-09T11:54:28.107013",
"tokens_seen": 100392960000,
"architecture": "Rose X1.5",
"features": [
"GQA",
"QK-Norm",
"RoPE",
"SwiGLU",
"RMSNorm",
"XSA",
"RefreshGate"
],
"optimizer": "muon_adamw"
}