ilsp/panellinies-exams-dataset

Dataset Card for Pan-Ex Dataset

Pan-Ex is a comprehensive dataset derived from publicly available exam questions and official solutions used for student admission to Higher Education Institutions in Greece via the Panhellenic Examinations (Πανελλαδικές Εξετάσεις).

The dataset includes questions with the following features:

• Subjects: Greek Language, Ancient Greek, History, Latin, Biology, Physics, Chemistry, Computer Science, Economics, and Mathematics.
• Educational Levels: Questions targeted at General Lyceum (Γενικό Λύκειο - GEL) graduates.
• Formats: Multiple Choice, True/False, Matching, Fill-in-the-Gaps, and Open-Ended questions.
• Modalities: Questions suitable for multimodal evaluation, featuring high-fidelity images/diagrams, LLM-generated image descriptions, and OCR transcriptions.

The benchmark can be used for the evaluation of LLMs on complex, multi-subject, multi-format questions in the Greek language. Additionally, it may be useful as a high-quality resource for quantitative educational research.

Dataset Creation

The source material was extracted from the official portals of the Greek Ministry of Education. It was then converted into the current format via specialized processing pipelines by the authors.

Disclaimer: While every effort has been made to ensure the accuracy and completeness of this structured dataset, any errors, omissions, or formatting issues are the result of the processing and transformation pipeline and are not related to the original source or the Ministry of Education.

Dataset Structure

Column	Description
id	Unique identifier (e.g., physics_gel_2024_B1).
subject	Academic subject (including physics, chemistry, ancient_greek).
format	multiple_choice, true_false, matching, fill_in_the_gaps, open_ended.
reference	Additional reference inputs: none, (text) passage, multimodal, table.
question	The core question text.
input	Additional context or passage required to answer the question.
images	Visual asset(s) (diagrams, photos, geometrical figures).
choices	Candidate answers for closed-ended questions.
answer_text	The correct answer/solution text.
answer_index	The index of the correct answer for multiple-choice questions.
image_description	LLM-generated textual descriptions of visual assets.
image_transcription	OCR/Text extraction from within the visual assets.
points	Assigned point value for the question (numeric). May be null if missing in the source.
year	The year of the exam.
school_type	The target education level/school type (e.g., gel).

Usage

import random

# Load a random sample
random_idx = random.randint(0, len(dataset) - 1)
sample = dataset[random_idx]

print(f"Sample Index: {random_idx} | ID: {sample['id']}")
print(f"Question: {sample['question']}\n")

# 1. Handle Multimodal Metadata (Descriptions & Transcriptions)
if sample.get('image_description'):
    print(f"🖼️  Image Description: {sample['image_description']}")

if sample.get('image_transcription'):
    print(f"📝 Image Transcription: {sample['image_transcription']}")

# 2. Handle Choices
if sample.get('choices'):
    print("\nChoices:")
    correct_idx = sample.get('answer_index')
    for i, choice in enumerate(sample['choices']):
        marker = "[✅]" if correct_idx is not None and i == int(correct_idx) else "[  ]"
        print(f"  {marker} {i}: {choice}")
    print(f"\nCorrect Answer Index: {correct_idx}")
else:
    print(f"\nAnswer: {sample['answer_text']}")

Known Data Gap

Holdout Set: The public Hugging Face repository purposefully excludes the 2026 exam questions and solutions. This specific year is utilized as a private holdout test set to ensure rigorous, contamination-free model evaluation.

Citation

If you make use of this dataset, please consider citing it as follows:

@misc{kyriazi2026passfailevaluatingllms,
      title={Pass or Fail? Evaluating LLMs on Two Greek Examination Benchmarks},
      author={Panagiota Kyriazi and Eleni Kasoura and Prokopis Prokopidis},
      year={2026},
      eprint={2609.34800},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2609.34800}
}

Dataset Card Contact

• Curated by: ILSP/Athena RC
• Language(s) (NLP): el
• License: cc-by-nc-4.0