Wrong distance result

[sup3rgiu]

Hi there,
First of all, thanks for the project! I found it very fast compared to other GPU implementations.
However, I think the Levenshtein distance is not calculated correctly.

For instance, consider this example:

tokenizer = Tokenizer()
padToken = 10
device = 'cuda:0'
ref = ['aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaacaccccccccccbbbbbbbbbbbbbb']
hyp = ['bbbbbbbbbbcbbbbbbcbccccbbbbccbbcbbcbccacaaaaaaaaaaaaaaaaaaaaaaaa']

ref = [tokenizer.tokenize(w) for w in ref]
hyp = [tokenizer.tokenize(w) for w in hyp]

# padding
x = pad_sequence(ref, batch_first=True, padding_value=padToken)
y = pad_sequence(hyp, batch_first=True, padding_value=padToken)

x, y = x.to(device), y.to(device)
pred = editdistance(x, y, padToken).to('cpu')
print(pred)

I get pred -> 49

However, it should be 62, as you can see when calculating the same distance using the famous Rapidfuzz Levensthein module:

from Levenshtein import distance
distance('bbbbbbbbbbcbbbbbbcbccccbbbbccbbcbbcbccacaaaaaaaaaaaaaaaaaaaaaaaa', 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaacaccccccccccbbbbbbbbbbbbbb')

or directly here.

Also, I think there is another problem with your implementation since the order of the arguments (i.e. strings) should not matter, yet running editdistance(y, x, padToken) returns 39.