dorsal/arxiv
View SchemaQalb: Largest State-of-the-Art Urdu Large Language Model for 230M Speakers with Systematic Continued Pre-training
| Authors | Muhammad Taimoor Hassan, Jawad Ahmed, Muhammad Awais |
|---|---|
| Categories | |
| ArXiv ID | 2601.08141vv1 |
| URL | https://arxiv.org/abs/2601.08141 |
| License | http://creativecommons.org/licenses/by-nc-sa/4.0/ |
Abstract
Despite remarkable progress in large language models, Urdu-a language spoken by over 230 million people-remains critically underrepresented in modern NLP systems. Existing multilingual models demonstrate poor performance on Urdu-specific tasks, struggling with the language's complex morphology, right-to-left Nastaliq script, and rich literary traditions. Even the base LLaMA-3.1 8B-Instruct model shows limited capability in generating fluent, contextually appropriate Urdu text. We introduce Qalb, an Urdu language model developed through a two-stage approach: continued pre-training followed by supervised fine-tuning. Starting from LLaMA 3.1 8B, we perform continued pre-training on a dataset of 1.97 billion tokens. This corpus comprises 1.84 billion tokens of diverse Urdu text-spanning news archives, classical and contemporary literature, government documents, and social media-combined with 140 million tokens of English Wikipedia data to prevent catastrophic forgetting. We then fine-tune the resulting model on the Alif Urdu-instruct dataset. Through extensive evaluation on Urdu-specific benchmarks, Qalb demonstrates substantial improvements, achieving a weighted average score of 90.34 and outperforming the previous state-of-the-art Alif-1.0-Instruct model (87.1) by 3.24 points, while also surpassing the base LLaMA-3.1 8B-Instruct model by 44.64 points. Qalb achieves state-of-the-art performance with comprehensive evaluation across seven diverse tasks including Classification, Sentiment Analysis, and Reasoning. Our results demonstrate that continued pre-training on diverse, high-quality language data, combined with targeted instruction fine-tuning, effectively adapts foundation models to low-resource languages.
{
"annotation_id": "cdc52fac-8b54-414b-ae97-9924b7041a7f",
"date_created": "2026-02-17T05:53:16.073000Z",
"date_modified": "2026-02-17T05:53:16.073000Z",
"file_hash": "7d4d4250847bae3c286f8a7abbdef97802c2a9133f40dddc670c5b798ed52115",
"private": false,
"record": {
"abstract": "Despite remarkable progress in large language models, Urdu-a language spoken by over 230 million people-remains critically underrepresented in modern NLP systems. Existing multilingual models demonstrate poor performance on Urdu-specific tasks, struggling with the language\u0027s complex morphology, right-to-left Nastaliq script, and rich literary traditions. Even the base LLaMA-3.1 8B-Instruct model shows limited capability in generating fluent, contextually appropriate Urdu text. We introduce Qalb, an Urdu language model developed through a two-stage approach: continued pre-training followed by supervised fine-tuning. Starting from LLaMA 3.1 8B, we perform continued pre-training on a dataset of 1.97 billion tokens. This corpus comprises 1.84 billion tokens of diverse Urdu text-spanning news archives, classical and contemporary literature, government documents, and social media-combined with 140 million tokens of English Wikipedia data to prevent catastrophic forgetting. We then fine-tune the resulting model on the Alif Urdu-instruct dataset. Through extensive evaluation on Urdu-specific benchmarks, Qalb demonstrates substantial improvements, achieving a weighted average score of 90.34 and outperforming the previous state-of-the-art Alif-1.0-Instruct model (87.1) by 3.24 points, while also surpassing the base LLaMA-3.1 8B-Instruct model by 44.64 points. Qalb achieves state-of-the-art performance with comprehensive evaluation across seven diverse tasks including Classification, Sentiment Analysis, and Reasoning. Our results demonstrate that continued pre-training on diverse, high-quality language data, combined with targeted instruction fine-tuning, effectively adapts foundation models to low-resource languages.",
"arxiv_id": "2601.08141",
"authors": [
"Muhammad Taimoor Hassan",
"Jawad Ahmed",
"Muhammad Awais"
],
"categories": [
"cs.CL",
"cs.AI",
"cs.LG"
],
"license": "http://creativecommons.org/licenses/by-nc-sa/4.0/",
"title": "Qalb: Largest State-of-the-Art Urdu Large Language Model for 230M Speakers with Systematic Continued Pre-training",
"url": "https://arxiv.org/abs/2601.08141",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "372f6227-5d61-4046-82be-98317adcaa8b",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}