deepseek-doc-and-code/train_results.json

{
    "epoch": 4.0,
    "train_loss": 0.4407963699874701,
    "train_runtime": 45928.5894,
    "train_samples": 29976,
    "train_samples_per_second": 2.611,
    "train_steps_per_second": 0.653
}