The article is a little light on details, but seems cool nonetheless. I’d like to see more details on the tensorflow modifications they made for parallel comput

The article is a little light on details, but seems cool nonetheless. I’d like to see more details on the tensorflow modifications they made for parallel computing. I’ve never had awesome results using distributed tf. The parameter server model just doesn’t seem to take good advantage of a really high performance interconnect.