DeepSpeed

b84287e8 - Add weight copy to base mlp ln, seperate norms from tensors in mlp/attn print statements