CultranAI at PalmX 2025: Data Augmentation for Cultural Knowledge Representation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Bhatti, Hunzalah Hassan, Ahmed, Youssef, Hasan, Md Arid, Alam, Firoj
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918152067088384
author Bhatti, Hunzalah Hassan
Ahmed, Youssef
Hasan, Md Arid
Alam, Firoj
author_facet Bhatti, Hunzalah Hassan
Ahmed, Youssef
Hasan, Md Arid
Alam, Firoj
contents In this paper, we report our participation to the PalmX cultural evaluation shared task. Our system, CultranAI, focused on data augmentation and LoRA fine-tuning of large language models (LLMs) for Arabic cultural knowledge representation. We benchmarked several LLMs to identify the best-performing model for the task. In addition to utilizing the PalmX dataset, we augmented it by incorporating the Palm dataset and curated a new dataset of over 22K culturally grounded multiple-choice questions (MCQs). Our experiments showed that the Fanar-1-9B-Instruct model achieved the highest performance. We fine-tuned this model on the combined augmented dataset of 22K+ MCQs. On the blind test set, our submitted system ranked 5th with an accuracy of 70.50%, while on the PalmX development set, it achieved an accuracy of 84.1%.
format Preprint
id arxiv_https___arxiv_org_abs_2508_17324
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CultranAI at PalmX 2025: Data Augmentation for Cultural Knowledge Representation
Bhatti, Hunzalah Hassan
Ahmed, Youssef
Hasan, Md Arid
Alam, Firoj
Computation and Language
Artificial Intelligence
68T50
F.2.2; I.2.7
In this paper, we report our participation to the PalmX cultural evaluation shared task. Our system, CultranAI, focused on data augmentation and LoRA fine-tuning of large language models (LLMs) for Arabic cultural knowledge representation. We benchmarked several LLMs to identify the best-performing model for the task. In addition to utilizing the PalmX dataset, we augmented it by incorporating the Palm dataset and curated a new dataset of over 22K culturally grounded multiple-choice questions (MCQs). Our experiments showed that the Fanar-1-9B-Instruct model achieved the highest performance. We fine-tuned this model on the combined augmented dataset of 22K+ MCQs. On the blind test set, our submitted system ranked 5th with an accuracy of 70.50%, while on the PalmX development set, it achieved an accuracy of 84.1%.
title CultranAI at PalmX 2025: Data Augmentation for Cultural Knowledge Representation
topic Computation and Language
Artificial Intelligence
68T50
F.2.2; I.2.7
url https://arxiv.org/abs/2508.17324