training loss is very low (0.177622) at the very beginning, is it normal?

Thanks for the nice work, I found the loss is very low from the start of training, is it normal?

below are my training log:

```
[INFO: logging.py: 67]: json_stats: {"_type": "train_iter", "cur_epoch": "1", "cur_iter": "10", "eta": "12:18:36", "loss": 0.177622, "lr": 0.100000, "mode": "train", "time_diff": 0.961633}

[INFO: logging.py: 67]: json_stats: {"_type": "train_iter", "cur_epoch": "1", "cur_iter": "20", "eta": "1 day, 14:35:48", "loss": 0.168755, "lr": 0.100000, "mode": "train", "time_diff": 3.015697}

[INFO: logging.py: 67]: json_stats: {"_type": "train_iter", "cur_epoch": "1", "cur_iter": "30", "eta": "12:21:22", "loss": 0.171540, "lr": 0.100000, "mode": "train", "time_diff": 0.965657}
```