# GemMaroc: Unlocking Darija Proficiency in LLMs with Minimal Data

arXiv: https://arxiv.org/abs/2505.17082 · HTML: https://mlsanity.com/paper/2505.17082/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 21 of its 36 references that are in the map.
- First posted on arXiv 20 May 2025.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 21 of 21)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [LIMA: Less Is More for Alignment](https://mlsanity.com/paper/2305.11206/) (2023)
- [Quantifying the Carbon Emissions of Machine Learning](https://mlsanity.com/paper/1910.09700/) (2019)
- [What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuning](https://mlsanity.com/paper/2312.15685/) (2023)
- [Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning](https://mlsanity.com/paper/2002.05651/) (2020)
- [Fanar: An Arabic-Centric Multimodal Generative AI Platform](https://mlsanity.com/paper/2501.13944/) (2025)
- [ALLaM: Large Language Models for Arabic and English](https://mlsanity.com/paper/2407.15390/) (2024)
- [xCoT: Cross-lingual Instruction Tuning for Cross-lingual Chain-of-Thought Reasoning](https://mlsanity.com/paper/2401.07037/) (2024)
- [X-Instruction: Aligning Language Model in Low-resource Languages with Self-curated Cross-lingual Instructions](https://mlsanity.com/paper/2405.19744/) (2024)
- [Mixture-of-Skills: Learning to Optimize Data Usage for Fine-Tuning Large Language Models](https://mlsanity.com/paper/2406.08811/) (2024)
- [Atlas-Chat: Adapting Large Language Models for Low-Resource Moroccan Arabic Dialect](https://mlsanity.com/paper/2409.17912/) (2024)
- [MURI: High-Quality Instruction Tuning Datasets for Low-Resource Languages via Reverse Instructions](https://mlsanity.com/paper/2409.12958/) (2024)
- [MIG: Automatic Data Selection for Instruction Tuning by Maximizing Information Gain in Semantic Space](https://mlsanity.com/paper/2504.13835/) (2025)
- [Investigating Multilingual Instruction-Tuning: Do Polyglot Models Demand for Multilingual Instructions?](https://mlsanity.com/paper/2402.13703/) (2024)
- [LIONs: An Empirically Optimized Approach to Align Language Models](https://mlsanity.com/paper/2407.06542/) (2024)
- [DEFT: Data Efficient Fine-Tuning for Pre-Trained Language Models via Unsupervised Core-Set Selection](https://mlsanity.com/paper/2310.16776/) (2023)
- [Boosting LLM Translation Skills without General Ability Loss via Rationale Distillation](https://mlsanity.com/paper/2410.13944/) (2024)
- [Fine-Tuning LLMs for Low-Resource Dialect Translation: The Case of Lebanese](https://mlsanity.com/paper/2505.00114/) (2025)

## Built on by (most cited first, 1 of 1)

- [Linear Semantic Segmentation for Low-Resource Spoken Dialects](https://mlsanity.com/paper/2605.06276/) (2026)

## Abstract (arXiv)

Open-source large language models (LLMs) still marginalise Moroccan Arabic (Darija), forcing practitioners either to bolt on heavyweight Arabic adapters or to sacrifice the very reasoning skills that make LLMs useful. We show that a rigorously quality-over-quantity alignment strategy can surface fluent Darija while safeguarding the backbone s cross-lingual reasoning at a sliver of the usual compute. We translate three compact instruction suites LIMA 1 K, DEITA 6 K and TULU 50 K into Darija, preserve 20 of the English originals, and add mathematics, coding and scientific prompts. A LoRA-tuned Gemma 3-4B trained on 5 K mixed instructions lifts DarijaMMLU from 32.8 to 42.7 ; adding the reasoning-dense TULU portion pushes it to 47.5 with no English regression. Scaling the identical recipe to Gemma 3-27B produces GemMaroc-27B, which matches Atlas-Chat on DarijaMMLU (61.6 ) and leaps ahead on Darija commonsense, scoring 60.5 on HellaSwag versus Atlas-Chat s 48.4 . Crucially, GemMaroc retains Gemma-27B s strong maths and general-reasoning ability, showing only minimal movement on GSM8K and English benchmarks. The entire model is trained in just 48 GPU.h, underscoring a Green AI pathway to inclusive, sustainable language technology. We release code, data and checkpoints to spur Darija-centric applications in education, public services and everyday digital interaction.