pub const THREE_DI_OFFSET: u32 = 125;Expand description
How far a 3Di token sits above its amino-acid counterpart.
ProstT5 extends this vocabulary with Foldseek’s 3Di alphabet as lowercase
letters at ids 128..=147, and every one of them is exactly its uppercase
residue’s id plus this offset — a = 128 = A(3) + 125, through
c = 147 = C(22) + 125. So there is no second table: 3Di is this table
shifted.
test_three_di_is_the_residue_table_shifted checks that against the ids
published in ProstT5’s added_tokens.json, so the shortcut is verified
rather than assumed.