{ "d_model": 256, "depth": 2, "dropout": 0.2, "partition": 0, "vocab_size": 106972, "weighted_loss": "none" }