model_40000/training_state.json

{
    "global_step": 80000,
    "update_step": 40000,
    "tokens_seen": 4000662612,
    "tokens_seen_before": 4000562384,
    "update_time": 2.4067161083221436
}