Hi @feizc! Thanks for your open-sourcing work. Just as the title, I am curious on how can I reproduce your FLOPs calculation (as reported in Table 1)?