My results in open-domain QA are much lower using the given checkpoint for CEPE-LLaMA-2-7B. Could you provide some insights into the potential causes for this decline?

[sunnynexus]

I'm curious about the discrepancies between my results (in red font) and the results presented in your paper (in black font), both obtained using the default parameters with the run_qa.sh script.

Could there be any potential errors on my end that could explain these differences?

[howard-yen]

Hi, thanks for your interest in our work. For CEPE at k = 10, we only use and put all the passages in the decoder model, which should match the results for LLaMA-2. There might have been a mistake in the config file, which I will look into. Are you also using the QA files from the google drive?

[sunnynexus]

Hi, thanks for your interest in our work. For CEPE at k = 10, we only use and put all the passages in the decoder model, which should match the results for LLaMA-2. There might have been a mistake in the config file, which I will look into. Are you also using the QA files from the google drive?

Thank you for your reply. Yes, I used the QA files from the google drive.

[sunnynexus]

I have tried running it multiple times, but the results are still not superior to the basic llama-2-7b model.