Issue: How to store a large distributed training model with KVStore?

What do you mean by “store”? Are you referring to training of the model or actually storing the parameters in a file? KVStore is used for updating parameters during training. I don’t quite understand what you want to achieve with KVStore.