Ketaba OCR LoRA

--- library_name: peft license: apache-2.0 language: - ar - en base_model: sherif1313/Arabic-English-handwritten-OCR-v3 pipeline_tag: image-to-text datasets: - NakbaNLP-2026 metrics: - cer - wer tags: - lora - qlora - dora - vision-language - manuscript - arabic - ocr - qwen2.5-vl model-index: - name: Ketaba-OCR results: - task: type: image-to-text name: Handwritten Text Recognition dataset: type: NakbaNLP-2026 name: Omar Al-Saleh Memoir Collection metrics: - type: cer value: 0.0819 - type: wer value: 0.2588 --- # Ketaba-OCR at AR-MS NakbaNLP 2026: Efficient Adaptation of Vision-Language Models for Handwritten Text Recognition <p align="center"> <img src="https://placehold.co/800x200/fef3c7/d97706?text=Ketab-Manuscript-OCR" alt="Arabic Manuscript OCR"> </p> This repository contains the official models and results for **Ketaba-OCR**, our submission to the **[NakbaNLP 2026 Shared Task](https://acrps.ai/nakba-nlp-manu-understanding-2026)** (AR-MS) on Arabic Manuscript Understanding (Subtask 2: Systems Track). We rank **1st on per-line evaluation** (CER 0.0819, WER 0.2588) and **3rd on the official (corpus-wide) leaderboard** (CER 0.0938, WER 0.2996). #### By: [Hassan Barmandah](https://scholar.google.com/citations?user=2VzOr0kAAAAJ&hl=en), [Fatimah Emad Eldin](https://scholar.google.com/citations?user=CfX6eA8AAAAJ&hl=ar), [Khloud Al Jallad](https://scholar.google.com/citations?user=A0EvL6cAAAAJ&hl=ar), [Omer Nacar](https://scholar.google.com/citations?user=pezf5FYAAAAJ&hl=en) — NAMAA Community (with Umm Al-Qura University, Trouve Labs, Syrian Society for Startups and Research, Tuwaiq Academy) [](https://github.com/HasanBGit/Ketab-OCR-LoRA) [](https://huggingface.co/collections/HassanB4/nakba-nlp-2026-arabic-manuscript-understanding-shared-task) [](https://colab.research.google.com/drive/1eKVDxrgMlyBLhDnW_0Gdm2kHC1153N7j?usp=sharing) [](LICENSE) --- ## Model Description This project introduces a **parameter-efficient approach** for Arabic handwritten text recognition (HTR) on historical manuscripts. The system is built upon Sherif's pretrained Ar...

Source context: 11 downloads · 5 likes · Pipeline image-to-text · Library peft · Repo HassanB4/Ketaba-OCR-LoRA