Downloads · 30 days
0
NAMAA-Space/aragenre-qwen2.5-7b-fewshot-v3
aragenre-qwen2.5-7b-fewshot-v3 is a text generation model from NAMAA-Space. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
A few-shot prompting package for Qwen/Qwen2.5-7B-Instruct, targeting hierarchical Arabic genre classification (broadgenre + specificgenre). No trained weights — this repo packages a prompt template and few-shot exampl…
Downloads · 30 days
0
Access
Public
Updated Aug 11, 2026
Repo size
—
Likes
0
Public
Click a slice to open those files.
.py24.3 KB · 69%
From the Hugging Face model README
A few-shot prompting package for Qwen/Qwen2.5-7B-Instruct, targeting hierarchical Arabic genre classification (broad_genre + specific_genre). No trained weights — this repo packages a prompt template and few-shot examples for local inference against the base model, not a fine-tuned checkpoint.
Authors: Hassan Barmandah (NAMAA Community; Umm Al-Qura University), Israa Elhosiny (NAMAA Community), Yousra El-Ghawi (NAMAA Community), Omer Nacar (NAMAA Community)
This prompt package's development-set score (0.9587 hierarchical F1) is not representative of real-world performance. Its few-shot examples were hand-picked from dev error analysis, specifically targeting hard Advice-vs-Motivational cases and implicit Religious Commentary examples that appeared in the dev set. Per the project's system-description paper, the broader fine-tuned/prompt-tuned lineage this belongs to collapsed to 0.22–0.44 hierarchical F1 on the actual 27,972-item hidden test set (74 specific genres under 6 broad genres) — dev-tuned prompting is subject to the same generalization gap as dev-tuned fine-tuning.
Contrast this with llm-zeroshot-reasoning (not in this release), which uses no dev-tuned few-shot examples at all and is a fairer estimate of true generalization for LLM prompting on this task.
The system that actually won for this team — 0.7013 hierarchical F1, 3rd of 18 teams on the official CodaBench leaderboard — was a separate, zero-shot DeepSeek-LLM pipeline with no fine-tuning and no dev-tuned few-shot examples (stage2_llm_zeroshot_pipeline/ in the project repo). This artifact is not that system. It is released here for transparency and reproducibility of the project's full experimental record, not as a recommended production classifier.
Qwen/Qwen2.5-7B-Instruct, few-shot prompting plus targeted disambiguation rules. Builds on an earlier iteration (dev hier_f1 0.8955) by fixing its 11 remaining errors with:
Qwen/Qwen2.5-7B-Instruct (local inference, no fine-tuning)
None — few-shot prompting only. Few-shot examples were selected using dev error analysis.
python qwen25_7b_fewshot_v3.py
Expects dev.json, dev_gold.json, dev_genre_definitions.json (obtain from the official AraGenre task registration).
If you use this work, please cite our system-description paper:
@inproceedings{barmandah-etal-2026-namaa,
title = {NAMAA at AraGenre 2026: From Encoder Baselines to Self-Consistent LLM Ensembling for Hierarchical Arabic Genre Classification},
author = {Barmandah, Hassan and Elhosiny, Israa and El-Ghawi, Yousra and Nacar, Omer},
booktitle = {Proceedings of the 4th Arabic Natural Language Processing Conference (ArabicNLP 2026)},
address = {Budapest, Hungary},
publisher = {Association for Computational Linguistics},
year = {2026},
}
Please also cite the AraGenre 2026 shared task overview paper:
@inproceedings{elhaj-etal-2026-aragenre,
title = {AraGenre 2026: A Hierarchical Definition-Guided Arabic Genre Classification Shared Task},
author = {El-Haj, Mo and Ezzini, Saad and Abudalfa, Shadi and Lamsiyah, Salima and Jarrar, Mustafa},
booktitle = {Proceedings of the 4th Arabic Natural Language Processing Conference (ArabicNLP 2026)},
address = {Budapest, Hungary},
publisher = {Association for Computational Linguistics},
year = {2026},
}
Apache 2.0