piiswrong commented on issue #8373: distribute training in fp16
URL: https://github.com/apache/incubator-mxnet/pull/8373#issuecomment-351214158
Closing due to outdated. @rahul003 please guide @solin319 to merge this into
the gradient compression framework.
piiswrong commented on issue #8373: distribute training in fp16
URL: https://github.com/apache/incubator-mxnet/pull/8373#issuecomment-339889212
My mistake, I meant general compression interface, not general n-bit
compression interface.
For now we can have 2bit and float16 after merging
piiswrong commented on issue #8373: distribute training in fp16
URL: https://github.com/apache/incubator-mxnet/pull/8373#issuecomment-338371778
@eric-haibin-lin @rahul003 I think this should be merged with the 2bit PR to
make a more general n-bit gradient compression feature.