experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix dtype check in GradientTape

Browse Source 
ResourceVariables handles will have Tensor type as dtype.resource.
Instead, we should check the dtype of the Variable Tensor.

PiperOrigin-RevId: 241623008
This commit is contained in:
Gaurav Jain 2019-04-02 16:04:19 -07:00 committed by TensorFlower Gardener
parent a6a0611a63
commit f44bcfbea9
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/python/eager/backprop.py
View File

@ -957,7 +957,7 @@ class GradientTape(object):
flat_sources = nest.flatten(sources)
flat_sources_raw = flat_sources
flat_sources = [_handle_or_self(x) for x in flat_sources]
for t in flat_sources:
for t in flat_sources_raw:
if not t.dtype.is_floating:
logging.vlog(
logging.WARN, "The dtype of the source tensor must be "