giannisdaras/ylg

ValueError: There are unused update ops

RenYurui opened this issue · 3 comments

RenYurui commented

Hi!
Thanks for your amazing repo!
I want to train your model. I have cloned your repo ("train" branch). But I got this error:

raise ValueError('There are unused update ops: %s' % unused_ops)
ValueError: There are unused update ops: {<tf.Operation 'replica_3/Generator/generator/g_block1/cbn_0/ema_update_ops' type=NoOp>,
<tf.Operation 'replica_2/Generator/generator/g_block2/cbn_1/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_1/Generator/generator/g_block4/cbn_1/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_2/Generator/generator/g_block3/cbn_0/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_2/Generator/generator/g_bn/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_3/Generator/generator/g_block1/cbn_1/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_1/Generator/generator/g_block5/cbn_0/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_2/Generator/generator/g_block3/cbn_1/ema_update_ops' type=NoOp>, 
<tf.Operation 'replica_1/Generator/generator/g_block5/cbn_1/ema_update_ops' type=NoOp>, 
......

Do you have any ideas about this?

My environments are:
CUDA 10.0
cudnn 7.4
tensorflow-gpu 2.0.1
tensorflow-gan 2.0.0

Thanks!

RenYurui commented

I follow the issue here to turn the checking flag to False. And the bug disappears.
However, I do not know whether this operation is correct.
Could you please give me some advice?

giannisdaras commented

Hi!
Thanks for bringing this to my attention.
So, you are trying to train this on GPU(s) environment right?
I have only run my experiments on TPUs so far.

RenYurui commented

Thanks for your reply!
I will close this issue cause the code seems correct when I turn the checking flag to False.
Updates will be provided if I find something new.

  • 🚀1