Reproduction Issue

[dayuyang1999]

Dear Authors,

thank you for your nice work and providing the code repository.

Can you please provide more detail on how to reproduce your results?

• can you please provide the checkpoint of the colBERT you used in your paper? becasue ColBERT did not have official released checkpoint. If you train your own checkpoint, could you provide a url toward the checkpoint?
• can you please provide the bash file you used for creating ColBERT FAISS index? I guess parameters to be used for indexing can influence search performance.
• and could you please also include some details about how you create indeces?

Thank you, Dayu

[littlewine]

Hi Dayu,

I emailed you the link to the checkpoint I used, since I don't have a good storage place rn. I can't recall the exact parameters I used for creating the FAISS index, I can check for any logs tonight/tomorrow, but no promises there.. My best guess is that I used what's described here. If you're unsure about any specific parameter you think is important, ask me in case I can recall/look it up.

PS: In case you are simply looking to compare/do analysis with ZeCo2 rankings and don't need a full reproduction, you can simply download the rankings provided in data/rankings/

[dayuyang1999]

Hi Author,

Thank you for your reply.

[littlewine]

I also just uploaded the checkpoint at HF modelhub: https://huggingface.co/uva-irlab/ZeCo2