Are model weights consistent at the initialization in distributed training?

Hello, are model weights consistent at the initialization in distributed training?
Suppose we have node1, node2, node3.
Are weigthts on these nodes are same?