model.outputs does not changed in svd compressor

[rafikg]

Hi @bachi55

keras_compressor/keras_compressor/compressor.py

Line 73 in 324ec9b

new_model = Model(model.inputs, model.outputs)

I debug the code and I see that the even the layers are swapped correctly but the model.output still have the old output not the new FactorizedDense layer

Model: "model_1"
....
....

_________________________________________________________________
dense_1 (FactorizedDense)    (None, 512)               193024    
_________________________________________________________________
batch_normalization_14 (Batc (None, 512)               2048      
_________________________________________________________________
dropout_10 (Dropout)         (None, 512)               0         
_________________________________________________________________
**dense_2 (FactorizedDense)    (None, 10)                4186**      
=================================================================
Total params: 15,001,418
Trainable params: 14,991,946
Non-trainable params: 9,472

but model.outputs still having the older dense_2 layer:

model.outputs
Out[4]: [<tf.Tensor 'dense_2/Softmax:0' shape=(?, 10) dtype=float32>]

Here the new model 
Model: new model
.......
......

dense_1 (FactorizedDense)    (None, 512)               193024    
_________________________________________________________________
batch_normalization_14 (Batc (None, 512)               2048      
_________________________________________________________________
dropout_10 (Dropout)         (None, 512)               0         
_________________________________________________________________
**dense_2 (Dense)              (None, 10)                5130**      
=================================================================
Total params: 14,931,786
Trainable params: 14,922,314
Non-trainable params: 9,472

---