DeepSpeed

69f3ce48 - Add weight copy to base mlp ln, seperate norms from tensors in mlp/attn print statements