eric-haibin-lin commented on a change in pull request #15124: [MXNET-1294]
Priority-based parameter propagation for improved data parallel training
throughput
URL: https://github.com/apache/incubator-mxnet/pull/15124#discussion_r370482473
##########
File path: python/mxnet/kvstore/base.py
##########
@@ -452,4 +457,13 @@ def create(name='local'):
from .kvstore import KVStore
kv = KVStore(handle)
set_kvstore_handle(kv.handle)
- return kv
+
Review comment:
Right, i meant `DMLC_PS_VAN_TYPE` as the env var.
Currently we have a few different communication backends: native, byteps,
horovod, p3. When changing the backend, users have to do one of the followings:
- `launch.py` + `gluon.Trainer(kv='dist')` for native
- `launch.py --horovod` + `gluon.Trainer(kv='horovod')`
- `launch.py --byteps` + `gluon.Trainer(kv='byteps')`
- `launch.py --p3` + `gluon.Trainer(kv='dist_p3')`
For users it should be fairly easy to follow. What do you think?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
