chunked/nltk_chunking/model_doc_reformer/chunk_25.txt

The premise is that the more "similar"
key query embedding vectors (in terms of cosine similarity) are to each other, the more likely they are assigned to
the same bucket.