llama3-8b-full-pretrain-junk-tweet-1m-en-reproduce-bs8/train_results.json

{
    "epoch": 3.0,
    "total_flos": 5863048888320.0,
    "train_loss": 2.8491841291374067,
    "train_runtime": 55958.28,
    "train_samples_per_second": 3.702,
    "train_steps_per_second": 0.463
}