fr bench pdf2md Benchmark [[📜 arXiv]](https://arxiv.org/abs/2602.11960) [[Dataset (🤗Hugging Face)]](https://huggingface.co/datasets/pulsia/fr bench pdf2md) [[pypi]](https://pypi.org/project/vlmparse/) [[vlmparse]](https://github.com/ld lab pulsia/vlmparse) [[Benchmark]](https://github.com/ld lab pulsia/benchpdf2md) [[Leaderboard]](https://huggingface.co/spaces/pulsia/fr bench pdf2md) fr bench pdf2md is a benchmark and dataset for evaluating PDF to Markdown conversion with vision–language models on challenging French documents. It is designed for practitioners who need reliable document parsing as a front end to RAG and other LLM pipelines, where the quality of the Markdown (structure + content) matters more than exact character level formatting. Inspired by the AllenAI OLMo OCR benchmark, fr bench pdf2md follows a unit test style evaluation: each page is associated with a small set of machine checkable tests that verify text presence/absence, reading order, and table structure. This makes failures easy to diagnose while avoiding over penalizing harmless formatting differences. The dataset focuses on difficult French pages selected from ~60k documents (CCPDF and Gallica) by compar…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy