How to disable change the output tensor name？

[zhen8838]

My model has multiple outputs when using model-optimization, the outputs node name has been changed like this:

982/982 [==============================] - 434s 442ms/step - loss: 17.4818 - prune_low_magnitude_l1_loss: 3.6453 - prune_low_magnitude_l2_loss: 10.2949 - prune_low_magnitude_l1_p: 0.8551 - prune_low_magnitude_l1_r: 0.6556 - prune_low_magnitude_l2_p: 0.7639 - prune_low_magnitude_l2_r: 0.4423 - val_loss: 16.9125 - val_prune_low_magnitude_l1_loss: 3.4677 - val_prune_low_magnitude_l2_loss: 9.9504 - val_prune_low_magnitude_l1_p: 0.8921 - val_prune_low_magnitude_l1_r: 0.7815 - val_prune_low_magnitude_l2_p: 0.8306 - val_prune_low_magnitude_l2_r: 0.5363

how to remove the prefix 'prune_low_magnitude' ?

[alanchiao]

Is l1_loss the name of one of your Keras layers? We intentionally prefixed different layer names to indicate they were pruned ("prune" would seem preferable to something as long as "prune_low_magnitude" though). I can try shortening the prefix to "prune".

If it's not a layer, it is a bit unexpected (we add the prefix here: https://github.com/tensorflow/model-optimization/blob/master/tensorflow_model_optimization/python/core/sparsity/keras/pruning_wrapper.py#L141). Could you share sample code for this l1_loss to reproduce?

[zhen8838]

The l1 is layer name, when I have multiple output layer, the loss name will change to l1_loss.

Can you just reduce the prefix length? Thanks

my code like ：

y1_reshape = Reshape((7, 10, anchor_num, 5 + class_num), name='l1')(y1)
y2_reshape = Reshape((14, 20, anchor_num, 5 + class_num), name='l2')(y2)

yolo_model_warpper = keras.Model(inputs, [y1_reshape, y2_reshape])
.
.
.
yolo_model_warpper.compile(loss=[loss_fn(),loss_fn()])