Feat: add OCR's muti-gpus and parallel processing support
What problem does this PR solve?
Add OCR's muti-gpus and parallel processing support
Type of change
- [x] New Feature (non-breaking change which adds functionality)
Looks generally good. However deepdoc OCR is not an obvious bottleneck. If it is, I would prefer running them with
triothreadpool. Refactoringchunk(rag/app/naive.py) function to async is not easy but amazing.
Hello, Thanks for your suggestions, I‘d like have a try.
@yuzhichang I've already switched the threadpool to Trio's. Please review it again. (The screenshot shows testing with non-continuous GPU settings.)
@debugdoctor Great job! The only issue is CI failure. Could you fix it?
@debugdoctor Great job! The only issue is CI failure. Could you fix it?
Thanks, let me fix it.^_^
@yuzhichang Checks passed, plz review it again.
@debugdoctor This PR passed CI and was merged. But I found PDF parsing is broken due to this PR. So I have reverted the merging. Please resolve my new comments, test and open another PR. Thanks!
Thanks for your review, I will fully test it before commit.