batch_size = 8, epoch = 1, learning_rate = 1e-4

Lora: r=16, lora_alpha=32, lora_dropout=0.05

metrics={'train_runtime': 729.5559, 'train_samples_per_second': 9.746, 'train_steps_per_second': 0.306, 'total_flos': 7.949170591137792e+16, 'train_loss': 2.817356810976037, 'epoch': 1.0}