Libya-based technology firm Smart Co for Technology Projects and Artificial Intelligence has announced the beta launch of LibiGPT, the country’s first national large language model (LLM). Developed by the company’s founder and CEO, Dr. Ali Othman Al-Baji, the model is engineered to address a critical gap in AI platforms by offering a deep understanding of Libyan culture, colloquial dialect, and regional terminologies.
The launch, which took place in Tripoli, was attended by key government figures, including the Minister of Economy and Trade, Mohamed Al-Hwej, signaling strong national support for the initiative.
Bridging a Critical AI Gap
Global models such as OpenAI’s ChatGPT and Google’s Gemini often lack the nuanced comprehension of local culture and the linguistic variations that characterize North African Arabic. LibiGPT was created specifically to overcome this challenge. By training on a corpus with a significant proportion of Arabic text, including Modern Standard Arabic (MSA) and the Libyan dialect (dārija), the model is designed to converse naturally and accurately in the local vernacular.
For Libyan enterprises, government institutions, and educational organizations, this offers a powerful tool with customization capabilities aligned with national history, culture, and regulatory requirements. Furthermore, all data is stored locally, addressing key data sovereignty and security concerns.
A Trio of Tailored Models
The developer has released three proprietary AI models to serve different needs across the ecosystem:
- LibiGPT-Base (7B): A 7-billion parameter model designed for lightweight applications.
- LibiGPT-Instruct (13B): A 13-billion parameter model optimized for powerful instruction-following tasks.
- LibiGPT-Enterprise (34B): A 34-billion parameter model intended for enterprise-level or sovereign-cloud deployments.
These versions allow for a range of applications, from public-facing chat interfaces to complex, secure deployments within national institutions.
Advanced Training on a Rich Arabic Corpus
LibiGPT was trained on a multi-hundred-billion-token corpus compiled from public and open-licensed Arabic datasets, MSA corpora, academic and journalistic texts, Arabic Wikipedia, and high-quality Creative Commons resources. The training data also included curated domain-specific corpora covering technical, legal, financial, and governmental documents from across the MENA region.
The development team employed a custom Arabic optimization pipeline that included dialect filtering, improved Arabic tokenization, and multi-stage toxicity and bias filtering. High-quality synthetic Arabic data was also generated to enhance the model’s robustness in handling complex reasoning tasks, long instructions, and code-switching between Arabic, English, and French.
Roadmap and Regional Context
The launch of LibiGPT is part of a growing trend across the Maghreb, where governments, commercial entities, and academic sectors are developing national AI models to better serve local populations.
According to Dr. Ali Al-Baji, the roadmap for LibiGPT includes expanding context windows to over 200,000 tokens, developing Arabic-domain expert models for sectors like law and healthcare, enhancing dialect understanding across the wider region, and releasing a formal technical report on its training methodology.
About Smart Co for Technology Projects and Artificial Intelligence
Smart Co for Technology Projects and Artificial Intelligence is a Libya-based technology company founded by Dr. Ali Othman Al-Baji. The company focuses on developing advanced AI solutions tailored to the linguistic and cultural needs of Libya and the broader Arabic-speaking world. Its flagship project, LibiGPT, is the first national large language model for Libya.
Source: Middle East AI News
