What breaks in document AI

Crashes and memory

13 of 13 quotes

• “i used Docling to process PDF. but Docling is heavy. and i have 8 GB ram.. so OOM issue occurred. pdf failed to process.”

• “after each document memory usage is increasing, till OOM and process is being killed.”

• “for some files it just fails to parse them properly and throws me this error: >Stage preprocess failed for run 1, pages \[66\]: std::bad\_alloc”

• “Processing 10 documents (200-400 pages each) reaches 60 GB peak RSS on CUDA. The memory is never reclaimed between documents.”

• “causes extreme RAM consumption — peaking at **37–60 GB** on a 63 GB server — making the service effectively unusable for large spreadsheets.”

• Hit this one too? We are building for it.

• “torch.OutOfMemoryError: CUDA out of memory.”

• “running `.load()` on a massive dataset (2GB+ of PDFs/Docs) is a guaranteed way to get an OOM (Out of Memory) error on a standard machine”

• “the memory occupied by mineru-api skyrockets.”

• “OOM issue leads to the killing of task executer.”

• “The description says the model run on 12GB cards.”

• “During the extraction process, **memory consumption increases significantly and does not seem to be released after a PDF is processed**.”

• “DocumentConverter.conver() method is consuming all available memory”

• “Now I have almost 62gb ram and it is still not enough? Its just a simple 5mb file.”

• Hit this one too? We are building for it.