Mulhem: Introducing the First LLM Trained Solely on Saudi Data Sets!

In a groundbreaking move, Riyadh- and Alkhobar-based Watad, a leading innovator in AI, cybersecurity, and smart city technologies, has unveiled “Mulhem,” the first-ever Saudi domain-specific Large Language Model (LLM) tailored exclusively to the rich linguistic and cultural tapestry of Saudi Arabia.

This pioneering bilingual Arabic/English AI model, developed with a meticulous focus on local relevance, was trained on an impressive 90 billion Arabic and 90 billion English data tokens sourced directly from the Kingdom.

Leveraging Watad’s carefully curated data sets, Mulhem stands as a testament to the company’s commitment to fostering AI solutions that are deeply rooted in the Saudi context and designed to cater to the unique needs and nuances of the region.

What is Mulhem?

Mulhem represents a significant leap in language model innovation, developed and trained by the visionary team at Watad Energy & Communications in Saudi Arabia. This large language model (LLM) stands out not only for its technical prowess, facilitated by NVIDIA’s high-performance computing systems, but also for its deep cultural resonance, as implied by its name, ‘Mulhem’ which means ‘inspiring’ in Arabic.

Mulhem is a proprietary technology, embodying Watad’s commitment to advancing AI in a manner that is both locally relevant and globally competitive. Despite its proprietary nature, Watad is considering paths to make Mulhem more accessible, including the possibility of open-source avenues, making it a beacon of inspiration and innovation in the AI realm.

How does Mulhem work?

At the core of Mulhem’s operational excellence is its bilingual capability, trained on an extensive foundation of 90 billion Arabic and 90 billion English data tokens. This dual-language fluency enables Mulhem to cater to a wide array of applications that demand proficiency in both Arabic and English.

The model’s training included a diverse dataset comprising over 70,000 Saudi-specific Q&A data points and more than 500,000 Arabic single and multi-turn conversations, preference datasets, and specialized data for context retrieval and offline data systems.

Such a comprehensive and Saudi-centric training regime ensures that Mulhem delivers responses and insights with unparalleled accuracy and relevance, particularly tuned to the Saudi and broader Arab contexts.

Developed on the principles of supervised fine-tuning and Direct Preference Optimization, Mulhem epitomizes the synergy of cutting-edge AI methodologies and the rich cultural tapestry of the Arab world, promising a new era of contextually aware and linguistically versatile AI applications.

In conclusion, Mulhem stands as a monumental achievement in the field of AI, seamlessly blending state-of-the-art technology with deep cultural understanding. Mulhem offers unparalleled linguistic versatility and cultural relevance, particularly in its dual-language Arabic and English capabilities. As Watad explores avenues for wider access, including potential open-source models, Mulhem is poised to not only serve the specific needs of the Saudi and Arab markets but also to inspire global advancements in AI.

Trump Supporters Manipulate Black Voters’ Perception with Fake AI Images