tensorflow: outputs indifferent to input when applying `merge_duplicate_nodes`

I think I might have located the root of #8698 It seems that merge_duplicate_nodes is the reason that quantize_nodes malfunctions. Whatever I do, whenever I apply a merge_duplicate_nodes somewhere during a graph transformation, the output becomes completely indifferent to the input.

(Unfortunately) I don’t get any error messages concerning this…

About this issue

  • Original URL
  • State: closed
  • Created 7 years ago
  • Comments: 22 (22 by maintainers)

Most upvoted comments

@Androbin I found an issue with dequantization on Android and I’m wondering if building with -DTENSORFLOW_DISABLE_META works for you. I’d appreciate if you would try it out. Thanks in advance!