llama-sft-muon/train_results.json

{
    "total_flos": 5.519394781646763e+19,
    "train_loss": 3.57993624316737,
    "train_runtime": 147576.5591,
    "train_samples": 1444084,
    "train_samples_per_second": 1.002,
    "train_steps_per_second": 0.125
}