Estimate total training FLOPs from parameter count and training tokens using Kaplan-style scaling laws, for compute planning.
N、D 單位需一致,結果數量級參考。