rahul003 commented on a change in pull request #11910: Improving documentation and error messages for Async distributed training with Gluon URL: https://github.com/apache/incubator-mxnet/pull/11910#discussion_r205923132
########## File path: docs/faq/distributed_training.md ########## @@ -73,6 +73,13 @@ These can be passed as arguments to the iterator. You can look at [example/gluon/image_classification.py](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py) to see an example usage. +### Updating weights +KVStore server supports two modes, one which aggregates the gradients and updates the weights using those gradients, and second where the server only aggregates gradients. In the latter case, when a worker process pulls from kvstore, it gets the aggregated gradients. The worker then uses these gradients and applies the weights locally. + +When using Gluon there is an option to choose between these modes by passing `update_on_kvstore` variable when you create the [Trainer](https://mxnet.incubator.apache.org/versions/master/api/python/gluon/gluon.html#mxnet.gluon.Trainer) object. Review comment: ok ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on 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
