Python 批处理上的Keras列车未更新网络

我正在尝试训练一个GAN模型，我正在使用train_on_batch更新网络，但似乎即使它计算了新的损耗，它也不会保存对网络的更改。我试着在训练前后打印损失，前后都一样。有什么问题

print(d_model_transfer.test_on_batch(Xsup_real, [ysup_real, label_sup_real]))
d_results = d_model_transfer.train_on_batch(Xsup_real, [ysup_real, label_sup_real])
print(d_results)
print(d_model_transfer.test_on_batch(Xsup_real, [ysup_real, label_sup_real]))

[0.08498648554086685, 0.0006484165205620229, 0.08433806896209717, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
[0.23932093381881714, 0.0010019546607509255, 0.23831897974014282, 1.0, 1.0, 1.0, 1.0, 0.875, 0.8749999403953552, 0.875, 0.875]

[0.08498648554086685, 0.0006484165205620229, 0.08433806896209717, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]