Closed kevin-keraudren closed 7 years ago
I explain the proposed fix in details here: https://github.com/tensorflow/tensorflow/pull/5599 and you can compare output of python train.py && python sample.py here: https://gist.github.com/kevin-keraudren/cbb948f06ac817d8efa5ed110b980fa9
python train.py && python sample.py
Without the fix, I attain a train_loss of 1.305 while in reaches 1.295 with the fix. More importantly, the output of the network sounds slightly more like Shakespeare.
train_loss
1.305
1.295
False alarm, closing the issue.
I explain the proposed fix in details here: https://github.com/tensorflow/tensorflow/pull/5599 and you can compare output of
python train.py && python sample.py
here: https://gist.github.com/kevin-keraudren/cbb948f06ac817d8efa5ed110b980fa9Without the fix, I attain a
train_loss
of1.305
while in reaches1.295
with the fix. More importantly, the output of the network sounds slightly more like Shakespeare.