Using device: cpu Loaded pretrained model. Epoch 0, Avg Loss: 3.1963 Epoch 1, Avg Loss: 2.4411 Epoch 2, Avg Loss: 1.4029 Epoch 3, Avg Loss: 0.5834 Epoch 4, Avg Loss: 0.2348 Epoch 5, Avg Loss: 0.1545 Epoch 6, Avg Loss: 0.1325 Epoch 7, Avg Loss: 0.1167 Epoch 8, Avg Loss: 0.1053 Epoch 9, Avg Loss: 0.0956 Model saved to models/sft_model.pt