Beir evaluation score using Pipeline.eval_beir is lower then using the one in Beir documentation

[huynguyengl99]

Hi all, (sorry for being a newbie), I have tested BM25Retriever and evaluated its corresponding Pipeline using the Pipeline.eval_beir but it seems the final score is lower than the one (my friend has tested) from the Beir documentation. Below is my code, did I miss any configuration or miscoding? Thank you for reading and it would be a favor if you can help me on that difference.
Furthermore, I have tested on the fiqa datasets.

Here are my code & evaluation:

Here is my friend's score following the Beir's guide

[vblagoje]

Hey @huynguyengl99, my guess would be to add the document_store to your query_pipeline?

[huynguyen-faty]

Hi @vblagoje , sorry for late response. I think the bm_25retriever already have the document_store inside because I need to pass the document store param when init it, so that we still have the evaluation score (if we dont have the document_store the score would be zero). Do you agree with me ?