OlmoCR Mix 2025 Benchmarkset
This dataset includes 10,000 PDF files extracted from a larger corpus, intended to assess the speed of OCR models and calibrate quantization. It is useful for OCR tasks on various documents.
Description
OlmoCR Mix 2025 Benchmarkset is a set of 10,000 PDF documents used primarily for OCR performance benchmarking and the calibration of quantified models. These files cover various types of documents and provide a basis for testing the robustness of OCR systems.
What is this dataset for?
- Benchmarking the speed and accuracy of OCR systems
- Quantization calibration for OCR models
- Evaluating models on a variety of PDF documents
Can it be enriched or improved?
The dataset can be enriched with additional manual annotations (corrected text extraction) to improve the quality of the benchmarks and allow supervised training tasks.
🔎 In summary
🧠 Recommended for
- OCR researchers
- QA engineers
- Quantified model developers
🔧 Compatible tools
- Tesseract
- EasyOCR
- PyTorch OCR pipelines
💡 Tip
Combine this dataset with corrected annotations for effective supervised training.
Frequently Asked Questions
Does this dataset contain corrected text annotations?
No, it is mainly used for benchmarking and calibration, with no corrected annotations provided.
Can this dataset be used to train an OCR model?
Not directly, except by adding quality manual or automatic annotations.
What is the main file format?
The dataset consists of PDF files representing various documents.




