import torch
import numpy as np
_ = torch.manual_seed(0)

d, k = 10, 10

# This way we can generate a rank-deficient matrix
W_rank = 2
W = torch.randn(d,W_rank) @ torch.randn(W_rank,k)
print(W)