IndexError: list index out of range after running a search on an already-created index. about ragatouille

[bclavie]

Hey, thank you for flagging! This was due to RAGatouille not exporting collection properly to disk, resulting in loading a far too short list when reloading the index! I've just pushed an updated version to PyPi where (experimentally) it should work properly, please let me know if you run into another issue!

[hochbergg]

I seem to still have this issue, on Mac with 0.0.4b1.
I indexed ~7k documents with the following code:

from ragatouille import RAGPretrainedModel
from ragatouille.data import CorpusProcessor
from tqdm import tqdm

from dao import Storage

if __name__ == '__main__':
    RAG = RAGPretrainedModel.from_pretrained("colbert-ir/colbertv2.0")
    db = Storage('db')
    processor = CorpusProcessor()

    documents = []
    for item in tqdm(db.get_all_items()):
        documents.append(json.loads(item[2])['text'])

    print(type(documents), type(documents[0])) # This outputs list, str
    processed = processor.process_corpus(documents)
    print(type(processed), type(processed[0])) # This outputs list, str

    index_name = RAG.index(index_name='rag_test', collection=processed)

    print('---')
    print(index_name)

Then I try querying with:

from ragatouille import RAGPretrainedModel

query = "What manga did Hayao Miyazaki write?"
RAG = RAGPretrainedModel.from_pretrained("colbert-ir/colbertv2.0")
results = RAG.search(query, index_name="colbert/indexes/rag_test")

print(results)

And get

  File "/Users/gal/Library/Caches/pypoetry/virtualenvs/knowledgedb-OLhc9Epa-py3.9/lib/python3.9/site-packages/colbert/data/collection.py", line 25, in __getitem__
    return self.data[item]
IndexError: list index out of range

With self.data being:

0 = {str} 'list with 7077 elements starting with...'
1 = {list: 3} [...]

[bclavie]

@hochbergg Thanks for flagging, this is a good catch!

This is due to me having missed an important update on the README. The way to load an existing index is no longer RAGPretrainedModel.from_pretrained() but RAGPretrainedModel.from_index(index_path).

I'm also hoping to restore support for the previous way, but am fiddling with the best way for the config to work for this. I'll update the README asap to clear up the confusion, let me know if it works for you!

[hochbergg]

Changed to

query = "What are some examples of adversarial AI attacks?"
RAG = RAGPretrainedModel.from_index(".ragatouille/colbert/indexes/rag_test")
results = RAG.search(query)

print(results)

And it works. Thank you. (Haven't seen the README update yet, but extrapolated from other example code)