AutoQueryEncoder not using query prefixes (for E5-like and other models)

[orionw]

Hi there!

Sorry, would make this a PR but I didn't have enough time to get around to it. Thanks again for adding prefixes so we could use models like E5. (issue #1720)

When using an E5-like model though, the code uses AutoQueryEncoder which ignores the prefix for the query. It's a pretty simple fix (a few lines). See https://github.com/castorini/pyserini/blob/master/pyserini/search/faiss/_searcher.py#L371

It should be like it is for some of the other classes (like how DkrrDprQueryEncoder handles it) if I understand correctly:

pyserini/pyserini/search/faiss/_searcher.py

Line 244 in 2bb342a

query = f'{self.prefix} {query}'

and

pyserini/pyserini/search/faiss/_searcher.py

Line 233 in 2bb342a

self.prefix = prefix

If I am misunderstanding which class E5 models go to, lmk, but I think they are supposed to go to AutoQueryEncoder.

[MXueguang]

Hi @orionw , in pyserini.search.faiss, we are currently using AutoQueryEncoder from pyserini.encode, rather than pyserini.search.faiss._searcher. The AutoQueryEncoder in _searcher is deprecated and would be removed.

https://github.com/castorini/pyserini/blob/2bb342acc124c69ec4fe13ebc3be0bd5a5bf497c/pyserini/search/faiss/__main__.py#L27C61-L27C77

[lintool]

Closed by #2008