Added del to save memory during reading

[jamesbraza]

Our readers are taking up too much RAM. This PR attempts to remove images from the readers early, to save overall RAM

---

Note

Reduces peak RAM during PDF parsing by freeing large intermediate image buffers as soon as they’re no longer needed.

• In paperqa_nemotron/reader.py: delete rendered_page after conversion, delete image_for_api post-API call, and delete cropped regions (region_pix) after saving
• In paperqa_pypdf/reader.py: delete pdfium bitmaps (pdfium_rendered_page, pix) after PIL conversion/saving for full-page screenshots, figure crops, and table crops
• Changes are localized to memory cleanup with no API or behavioral surface changes

^{Written by Cursor Bugbot for commit db16c88. Configure here.}

[dosubot]

Related Documentation

Checked 1 published document(s) in 1 knowledge base(s). No updates required.

^{How did I do? Any feedback?}  

[Copilot]

Pull request overview

This PR adds explicit del statements to free image objects earlier in the PDF reading process to reduce RAM usage. The changes target image-related objects (pdfium bitmaps, PIL images, and numpy arrays) that are no longer needed after extraction or rendering.

Key Changes:

• Added del statements to free pdfium bitmap objects immediately after they're converted or used
• Added del for intermediate numpy array used for API calls once processing is complete
• Added del for cropped PIL images after they're saved to BytesIO buffers

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
packages/paper-qa-pypdf/src/paperqa_pypdf/reader.py	Frees pdfium bitmap memory for full-page renders, clustered image regions, and table regions immediately after conversion to PNG format
packages/paper-qa-nemotron/src/paperqa_nemotron/reader.py	Frees pdfium bitmap memory after PIL conversion, numpy array memory after API calls complete, and cropped image memory after saving to buffer

After thorough review, all the deletions are safe and correctly placed. The variables are deleted only after their last usage, and any derived objects (like PIL images created from pdfium bitmaps) are retained as needed. The comments accurately describe the memory being freed. This is a clean optimization that should help reduce memory usage during PDF processing without introducing any bugs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.