how to calculate Neff

I'm trying to replicate how Neff was calculated for the 2021 AlphaFold paper. It cites Wu et al for their description of Neff, but I find that a bit confusing. Here's a screenshot of their description:

I understand that to get Neff, you iterate over every sequence i in the MSA, and sum 1 divided by weight_i for each sequence i. Weight_i is in turn the sum of weight_ij, which involves going over all the other sequences j and comparing them to i, seeing how similar that sequence is to the others. However, what exactly weight_ij is isn't clearly stated. Is it perhaps 1 if the sequence identity is >= 62%, and then the sequence identity (e.g 0.5 for 50% sequence identity) otherwise? Or would it be 0 if the sequence identity is below 62%?

Beyond this, AlphaFold seems to be modifying the original Neff cited in two ways: 1) by using an 80% cutoff instead of 62%, and 2) by computing on a per residue basis. So would replicating Neff as used in the 2021 paper involve following my description given above, except with an 80% sequence identity cutoff, and if there's a gap weight_ij becomes 0? This is my impression, but I've realized it's based on several assumptions, so I should check it.

Additionally, could you point me to what algorithm / code AlphaFold used to calculate sequence identity?

Thank you so much for your time.