Closed hasnain095 closed 1 year ago
Dear @hasnain095,
It is OOM error and means you ran out of memory. But if it only occurs with -di flag, memory leakage can be a possible reason. By the way this will be checked as soon as possible.
Dear @hasnain095 ,
As you can see in my case all images were processed without any error and I could reproduce the same OOM error.
And here my GPU specifications which is used for the test (8 GB)
I too could not reproduce the OOM error anymore with a 8 GB GPU and using the -di
flag, so will close here now.
When eynollah is run with -di flag, in some instances it crashes and stops processing.
The command used:
eynollah -di ./issue_images/000___2020___sustainability_report___ed8c9421412e74c2ad60da1f702dafa2__174456655 -m ./models/ -o ./page_xml/sustainability_report/2020/ed8c9421412e74c2ad60da1f702dafa2__174456655/ -light -fl -ipe -tab
The images were generated from pdfs using pdf2image python package.
Server specifications are as follows:
Server | GPUs| CPUs| RAM| GPU Memory g3s.xlarge | 1 | 4 | 30.5 | 8
Tesla M60 GPU
I've attached the images of one pdfs, and also attached 10 sample pdfs that are causing the issue. Please note that some of the documents are not in English, would that cause any issue? 0002020sustainability_report_ed8c9421412e74c2ad60da1f702dafa2174456655.zip pdfs
It was recommended earlier to use -ipe flag to avoid another issue occurring: https://github.com/qurator-spk/eynollah/issues/77
The issue does not appear when running with the -i flag. But because we have a lot of documents to process, we need the -di flag.
The stack trace is: