Fused conv implementation does not support grouped convolutions for now.

最近在用tensorflow框架自己搭网络，有两层卷积层想使用分组卷积，但是运行就会报错“Fused conv implementation does not support grouped convolutions for now.”，把groups改成1就不会报这个错误了，所以想请教一下如何解决这个问题，以下是我分组卷积的代码：

conv3= Conv2D(64, 5, activation='relu', padding='same',groups=2,
                        kernel_initializer=keras.initializers.TruncatedNormal(stddev=0.01),
                        bias_initializer=keras.initializers.constant(0.01))(conv2)
conv4 = Conv2D(32, 5, activation='relu', padding='same',groups=2,
                        kernel_initializer=keras.initializers.TruncatedNormal(stddev=0.01),
                        bias_initializer=keras.initializers.constant(0.01))(conv3)