yg-dev-git
yg-dev-git
The name of the function should be "operator
> You can use `all_gather` to gather features together. The caveat is that you need to manually propagate gradients through `all_gather` op, as it doesn't auto-bp. I finally make it...
> Hi, can you share your code about how to implement this? I am not familiar with `all_gather` .etc operations. Thanks a lot. First, install [diffdist](https://github.com/ag14774/diffdist). Then put the following...
Yes, but I'm not sure this is bug-free. On Tue, Jul 21, 2020, 5:27 PM XU Shijian wrote: > Thank you for your quick reply. > > So, then I...