Neal Wu comments

Repositories
Issues
Comments

Results 3 comments of


                                            Neal Wu

Using batch_size * num_gpus as batch_size in exponential_decay calculation?

For multi-GPU CIFAR-10 we [recommend](https://github.com/tensorflow/models/blob/master/tutorials/image/cifar10/README.md) looking at https://github.com/tensorflow/models/tree/master/tutorials/image/cifar10_estimator instead. The other implementation you linked is likely out of date.

Extension no longer available?

Yes, unfortunately Facebook took it down for logo infringement 😔

[Explanation/Clarification] radix_sort.cc

#1 and #3 are just situations where we would rather do comparison sort. #2 is to smooth out the number of bits for each pass; e.g., instead of 10 bits,...