Translation_server : bad size of the attn matrix?

In the translation_server json reply, it seems that the attn matrix have one more line than the number of tokens of the translation sentence built, while the number of columns are equal to the number of tokens of the source sentence. Is there an explanation for this ? Is this a bug ? Is there something wrong on my side ?

PS : is this because of the </s> tag ?

Yep, that’s a bug:


1 Like