Ro Sft Finepdfs: Romanian PDF Page Images and Extracted Text for VLM Training

A Romanian-language subset of the FinePDFs corpus, prepared for OCR, containing pairs of page images and extracted text. The dataset is part of an instruction fine-tuning protocol for Romanian Vision-Language Models proposed in the paper 'Înțelegi românește?'. It was created by OpenLLM-Ro and last updated on June 5, 2026.

Use Cases

• Train Romanian vision-language models based on image-text pairs.
• Fine-tune OCR models for Romanian text extraction from PDFs.
• Benchmark multimodal learning techniques on a non-English language.
• Develop instruction-following AI agents for Romanian visual content.

Strengths

• Derived from FinePDFs, described as the largest publicly available corpus sourced exclusively from PDFs.
• Specifically prepared for OCR tasks with aligned image-text pairs.
• Part of a documented research protocol for Romanian VLMs.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The specific size, file formats, and license for this Romanian split are unknown.

Provenance

Source

OpenLLM-Ro, derived from the FinePDFs corpus.

Collection Method

Likely extracted and processed from PDF documents for OCR.

Freshness

Last updated 2026-06-05 06:05:11; freshness should be verified.

Geography

Romanian language focus.