Greg Pauloski

Has there been any progress on integrating this into main? We are running into the same issue with the count of many two-character tokens overflowing and not ending up in...

Thanks, @Narsil! I'll try `spm`.

@imazik @chanwit Is this still being worked on? I would like to work on it as well.

My group focused on some other open issues so we did not end up doing any work on this.

Can also use `allgather_coalesced` instead of gradient/inverse broadcast.