mpapucci's picture
re-trained mutlitask model to align it with the singletask ones
ed02c05
{
"epoch": 2.0,
"total_flos": 1.0506418329673728e+16,
"train_loss": 1.2627147819042441,
"train_runtime": 2692.7589,
"train_samples": 32526,
"train_samples_per_second": 24.158,
"train_steps_per_second": 3.02
}