When I use the cal_logp of all dataset,I met the question.This

RLHF-V / RLAIF-V

RLAIF-V: Aligning MLLMs through Open-Source AI Feedback for Super GPT-4V Trustworthiness

200 stars 6 forks source link

When I use the cal_logp of all dataset,I met the question.This #20

Closed XiaoLei2123 closed 2 weeks ago

XiaoLei2123 commented 1 month ago

img_v3_02do_b346611a-ead2-4f84-9dd9-2b74cdc77afg

Haoye17 commented 1 month ago

Hi @XiaoLei2123, thanks for your interest in our work~

This error may occur when the size of a single parquet file is larger than 2GB. To deal with it, we previously updated our code to save the final logp parquets in every 5000 samples, could you please update the code base to see if it still occurs?

If there are further questions, we are willing to help!

XiaoLei2123 commented 4 weeks ago

Thank you for your reply.I will update my code to solve this question.

XiaoLei2123 commented 4 weeks ago

If i want to solve this question , which python files I need to update?Can you offer a list about this question.Thank you!

Haoye17 commented 3 weeks ago

Hi @XiaoLei2123 ! The files you need to update are:

muffin/data/datasets.py: https://github.com/RLHF-V/RLAIF-V/blob/main/muffin/data/datasets.py
muffin/eval/muffin_inference_logp.py: https://github.com/RLHF-V/RLAIF-V/blob/main/muffin/eval/muffin_inference_logp.py

You can try to see if this can help~