Optimization difference of PyTorch and Tensorflow

[whq-hqw]

Hi, thanks for the benchmark.

I did a test on PyTorch vs Keras with Tensorflow backend, the dataset was Cifar-10, both of them started from completely same initialization. I found no matter how I tried, the optimization result of PyTorch was always a subversion of Keras(Tensorflow)'s result. I just wonder if you have noticed the same problem.

[tango4j]

Hi. I think tensorflow will be still faster than pytorch while results vary over dataset and architecture. You might want to test with various examples. If you are not using eager excution, TF will be faster in general. It will be awesome if you can share the benchmark result. Thanks. 2019년 8월 26일 (월) 오전 4:25, Wang Hanqin <notifications@github.com>님이 작성:

…

Hi, thanks for the benchmark. I did a test on PyTorch vs Keras with Tensorflow backend, the dataset was Cifar-10, both of them started from completely same initialization. I found no matter how I tried, the optimization result of PyTorch was always a subversion of Keras(Tensorflow)'s result. I just wonder if you have noticed the same problem. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1?email_source=notifications&email_token=ADEZOW5ZKHF5LKKD3QOJQWLQGO4TXA5CNFSM4IPOUOSKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HHLLLBQ>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADEZOW4KM6RFQDI2J3QICD3QGO4TXANCNFSM4IPOUOSA> .