A question about esm-1v

[lzygitk7]

Hello, ESM is a great job. I have a question about esm-1v 'masked-marginals' strategy.
for the command;
all_token_probs.append(token_probs[:, i]) # vocab size
from my understanding, the token_probs has a shape of (n, 33), where n is the batch_tokens.size(1) as well as the length of the protein. So I am really confused about token_probs[:, i], should it be token_probs[i, :] instead?
And esm-1v only fits the situation of a single mutation, for other mutations, are there any other score strategies?
Thank you in advance!