Implement of lbfgs function with PyTorch

[zhmzm]

Hi,
Your paper is really impressive. I encounter some difficulties when I implement the function lbfgs in PyTorch. In Algorithm 1, if there are n parameters in the Network, the outcome of line 1 is n * n. In ResNet18 with BN layer n is 11173962, how can I store such a large matrix in memory? Or do I misunderstand anything? Thanks for your reply.

[zhmzm]

The issue is fixed. The weights are recorded as weight[11173962][10], so actually, the matrix is 10*10.