Kazakh Language Gets its National LLM with a Groundbreaking Partnership of Kazakh Research Institutions and VEON’s QazCode
VEON announces the launch of an open-source Kazakh-language large language model (Kaz-LLM), developed through a consortium led by Kazakhstan's Ministry of Digital, Innovations, and Aerospace Industry. The model, featuring 150 billion tokens and versions with 8 billion and 70 billion parameters, supports Kazakh, Turkish, English, and Russian languages.
The project involves collaboration between the Institute of Smart Systems and Artificial Intelligence at Nazarbayev University, VEON's QazCode, Beeline Kazakhstan, and the Astana Hub. The model has been published on the Hugging Face platform, marking a significant step in developing AI-powered services in Kazakhstan. This initiative builds on VEON's previous success with the Kaz-RoBERTA-conversational model, which has been downloaded over 3,000 times and is currently used in Beeline Kazakhstan's customer service.
VEON annuncia il lancio di un modello linguistico di grandi dimensioni in lingua kazaka open-source (Kaz-LLM), sviluppato attraverso un consorzio guidato dal Ministero delle Digitalizzazioni, Innovazioni e Industria Aerospaziale del Kazakistan. Il modello, che dispone di 150 miliardi di token e versioni con 8 miliardi e 70 miliardi di parametri, supporta le lingue kazaka, turca, inglese e russa.
Il progetto prevede la collaborazione tra l'Istituto di Sistemi Intelligenti e Intelligenza Artificiale dell'Università Nazarbayev, QazCode di VEON, Beeline Kazakhstan e Astana Hub. Il modello è stato pubblicato sulla piattaforma Hugging Face, segnando un passo significativo nello sviluppo di servizi alimentati da IA in Kazakistan. Questa iniziativa si basa sul successo precedente di VEON con il modello conversazionale Kaz-RoBERTA, che è stato scaricato oltre 3.000 volte ed è attualmente utilizzato nel servizio clienti di Beeline Kazakhstan.
VEON anuncia el lanzamiento de un modelo de lenguaje grande en idioma kazajo de código abierto (Kaz-LLM), desarrollado a través de un consorcio liderado por el Ministerio de Digitalización, Innovaciones e Industria Aeroespacial de Kazajistán. El modelo, que cuenta con 150 mil millones de tokens y versiones con 8 mil millones y 70 mil millones de parámetros, admite los idiomas kazajo, turco, inglés y ruso.
El proyecto implica la colaboración entre el Instituto de Sistemas Inteligentes e Inteligencia Artificial de la Universidad Nazarbayev, QazCode de VEON, Beeline Kazajistán y Astana Hub. El modelo ha sido publicado en la plataforma Hugging Face, marcando un paso significativo en el desarrollo de servicios impulsados por IA en Kazajistán. Esta iniciativa se basa en el éxito previo de VEON con el modelo conversacional Kaz-RoBERTA, que ha sido descargado más de 3,000 veces y actualmente se utiliza en el servicio de atención al cliente de Beeline Kazajistán.
VEON은 카자흐어 오픈 소스 대형 언어 모델(Kaz-LLM)의 출시를 발표했습니다. 이 모델은 카자흐스탄의 디지털화, 혁신 및 항공 우주 산업부가 주도하는 컨소시엄을 통해 개발되었습니다. 이 모델은 1,500억 개 토큰과 각각 80억 개와 700억 개의 파라미터를 가진 버전을 포함하여 카자흐어, 터키어, 영어 및 러시아어를 지원합니다.
이 프로젝트는 나자르바예프 대학교의 스마트 시스템 및 인공지능 연구소, VEON의 QazCode, Beeline 카자흐스탄 및 아스타나 허브 간의 협력을 포함합니다. 이 모델은 Hugging Face 플랫폼에 게시되어 카자흐스탄에서 AI 기반 서비스 개발의 중요한 이정표가 되었습니다. 이 이니셔티브는 VEON이 이전에 성공을 거둔 Kaz-RoBERTA 대화 모델을 기반으로 하며, 이 모델은 3,000회 이상 다운로드되었으며 현재 Beeline 카자흐스탄의 고객 서비스에 사용되고 있습니다.
VEON annonce le lancement d'un modèle de langage de grande taille en langue kazakhe en open source (Kaz-LLM), développé par un consortium dirigé par le ministère de la Numérisation, des Innovations et de l'Industrie Aérospatiale du Kazakhstan. Le modèle, comportant 150 milliards de tokens et des versions avec 8 milliards et 70 milliards de paramètres, prend en charge les langues kazakhe, turque, anglaise et russe.
Le projet implique une collaboration entre l'Institut des systèmes intelligents et de l'intelligence artificielle de l'Université Nazarbayev, QazCode de VEON, Beeline Kazakhstan et l'Astana Hub. Le modèle a été publié sur la plateforme Hugging Face, marquant une étape significative dans le développement de services alimentés par l'intelligence artificielle au Kazakhstan. Cette initiative s'appuie sur le précédent succès de VEON avec le modèle conversationnel Kaz-RoBERTA, qui a été téléchargé plus de 3 000 fois et est actuellement utilisé dans le service client de Beeline Kazakhstan.
VEON kündigt den Start eines offenen, kasachischsprachigen großen Sprachmodells (Kaz-LLM) an, das durch ein Konsortium unter der Leitung des Ministeriums für Digitalisierung, Innovationen und der Luft- und Raumfahrtindustrie Kasachstans entwickelt wurde. Das Modell, das über 150 Milliarden Tokens verfügt und Versionen mit 8 Milliarden und 70 Milliarden Parametern enthält, unterstützt die Sprachen Kasachisch, Türkisch, Englisch und Russisch.
Das Projekt umfasst die Zusammenarbeit zwischen dem Institut für intelligente Systeme und künstliche Intelligenz der Nazarbayev-Universität, VEONs QazCode, Beeline Kasachstan und dem Astana Hub. Das Modell wurde auf der Plattform Hugging Face veröffentlicht, was einen bedeutenden Schritt in der Entwicklung von KI-unterstützten Dienstleistungen in Kasachstan darstellt. Diese Initiative baut auf dem bisherigen Erfolg von VEON mit dem konversationellen Modell Kaz-RoBERTA auf, das über 3.000 Mal heruntergeladen wurde und derzeit im Kundenservice von Beeline Kasachstan verwendet wird.
- First major Kazakh-language LLM development positions VEON as a leader in low-resource language AI
- Strategic partnership with government and research institutions strengthens market position in Kazakhstan
- Previous AI model (Kaz-RoBERTA) successfully implemented in customer service operations
- Expansion of AI capabilities could lead to new revenue streams and improved service delivery
- None.
Insights
The launch of Kaz-LLM represents a significant technological advancement in natural language processing for Kazakhstan, with 150 billion tokens and versions featuring 8 billion and 70 billion parameters. This development is particularly noteworthy as it addresses the critical gap in AI language models for low-resource languages. The model's multilingual capabilities across Kazakh, Turkish, English and Russian position it as a versatile tool for developers and businesses in the region.
The open-source nature of the model, published on Hugging Face, allows for widespread adoption and further development by the tech community. VEON's previous success with Kaz-RoBERTA-conversational, which has seen over 3,000 downloads, demonstrates the market demand for such solutions. The collaboration with GSMA Foundry and Barcelona Supercomputing Center suggests strong potential for knowledge transfer and future improvements.
For VEON, this initiative strengthens its market position in Kazakhstan, its most advanced market for AI capabilities. The partnership with government institutions and research bodies enhances VEON's credibility in the AI space and could lead to increased adoption of its digital services. The development aligns with the growing trend of localized AI solutions, potentially opening new revenue streams through AI-powered products and services.
The focus on low-resource languages is strategically significant, as it addresses an underserved market segment with potential access to half a billion people in VEON's markets. This positions VEON favorably for expansion into other markets with similar linguistic challenges. The success of their previous Kaz-RoBERTA-conversational model in customer service applications demonstrates clear commercial viability.
Dubai, Amsterdam and Astana, 11 December 2024: VEON Ltd. (Nasdaq: VEON), a global digital operator, is pleased to note the launch of an open-source Kazakh-language large language model (Kaz-LLM), developed by a consortium coordinated by the Ministry of Digital, Innovations, and Aerospace Industry of the Republic of Kazakhstan. The development of Kaz-LLM was led by the Institute of Smart Systems and Artificial Intelligence at the Nazarbayev University (ISSAI NU) of Kazakhstan, in partnership with VEON’s QazCode, Beeline Kazakhstan and the Astana Hub.
With over 150 billion tokens collected, curated, synthesized and translated, the Kaz-LLM is capable of interacting in Kazakh language as well as in Turkish, English and Russian. With an 8 billion and a 70 billion parameter versions, the Kaz-LLM, developed in Kazakhstan, will help accelerate the creation and adoption of AI-powered products and services in the country. The model has been published on the Hugging Face platform for developers, ahead of its full launch.
The initiative, in which VEON’s QazCode is the only private sector partner, aligns closely with VEON’s mission to provide speakers of low-resource languages with augmented intelligence tools to enhance their daily lives, starting with Kazakhstan, VEON’s most advanced market in terms of augmented intelligence capabilities.
“The launch of the open-source Kaz-LLM represents a pivotal step forward in the development of Kazakhstan’s AI ecosystem. This initiative reflects our unwavering commitment to fostering innovation and advancing scientific endeavors that drive technological progress. I am confident that this groundbreaking model will help bridge the digital divide, bringing accessible and inclusive digital services to every Kazakhstani, regardless of their native language,” said Zhaslan Madiyev, Minister of Digital, Innovations & Aerospace of the Republic of Kazakhstan.
“AI, augmented Intelligence, has immense potential to amplify and augment human skills and capabilities; empowering doctors to deliver better care, teachers to inspire deeper learning, farmers to optimize yields, and students to excel. Yet, AI has a bias towards high-resource languages with greater digital representation. Operating in markets where national languages are not rich in digital research libraries, VEON is uniquely positioned to bridge this linguistic gap, ensuring that the half a billion people in our markets enjoy equal opportunities in the digital age. I warmly congratulate the Republic of Kazakhstan on the launch of Kaz-LLM. It is a privilege to have contributed to this landmark achievement through Beeline Kazakhstan and QazCode. We look forward to sharing these learnings globally and advancing inclusivity in AI,” said Kaan Terzioglu, CEO of VEON Group.
"We are delighted to have partnered with Kazakhstan’s leading research institutions for the development of Kaz-LLM. Our data science professionals and developers have brought in all the experience of Beeline Kazakhstan and QazCode in developing AI-based products into this joint national project. This major undertaking will benefit the entire digital ecosystem of Kazakhstan, ensuring that the country is among the leaders of augmented intelligence. It will also be a first for us at VEON Group, in line with our focus on addressing the AI language gap for the benefit of billions of speakers of low-resource languages,” said Alexey Sharavar, CEO of QazCode.
Beeline Kazakhstan and QazCode have already launched several AI products developed in-house, including the open-source LLM Kaz-RoBERTA-conversational model, which was the first Kazakh-language AI model with 2 billion parameters. Kaz-RoBERTA-conversational is currently used for customer service interactions on Beeline Kazakhstan’s digital platforms and is an open-source model that has been downloaded over three thousand times on the Hugging Face platform.
Beeline Kazakhstan and QazCode are also actively involved in contributing to international know-how on LLM development for low-resource languages. QazCode cooperates with the GSMA Foundry and the Barcelona Supercomputing Center, which has developed an LLM for the Catalan language, for sharing of expertise on LLM development.
About ISSAI
Institute of Smart Systems and Artificial Intelligence (ISSAI) was founded in September 2019 to serve as the driver of research and innovation in the digital sphere of Kazakhstan with the focus on AI research. ISSAI provides an agile framework for research, innovation and collaboration with national and international partners in education, industry and government and contributes to the digital ecosystem of Kazakhstan in the advancement of national development goals.
About Beeline Kazakhstan and QazCode
Beeline Kazakhstan serves 11 million customers with mobile connectivity and two million with fixed internet services. Since 2018, the company has been executing its digital operator strategy. Over the past five years, leveraging its expertise in digital solution development, Beeline has created an ecosystem of 60 internal and external products, and serves a total monthly active user base of 11.6 million with its digital products as of June 2024. Beeline Kazakhstan is majority-owned by VEON.
QazCode, the software development company of Beeline Kazakhstan, is one of the largest companies in Kazakhstan with a team of 700 people including 350 developers. QazCode creates solutions such as telecoms process automation, gamification, entertainment and IT productivity using an artificial intelligence approach.
About VEON
VEON is a Nasdaq-listed digital operator that provides converged connectivity and digital services to nearly 160 million customers. Operating across six countries that are home to more than
Disclaimer
This release contains “forward-looking statements,” as the phrase is defined in Section 27A of the U.S. Securities Act of 1933, as amended, and Section 21E of the U.S. Securities Exchange Act of 1934, as amended. Forward-looking statements are not historical facts, and include statements relating to, among other things, VEON’s strategy regarding development of AI products and capabilities. Forward-looking statements are inherently subject to risks and uncertainties, many of which VEON cannot predict with accuracy and some of which VEON might not even anticipate. The forward-looking statements contained in this release speak only as of the date of this release. VEON does not undertake to publicly update, except as required by U.S. federal securities laws, any forward-looking statement to reflect events or circumstances after such dates or to reflect the occurrence of unanticipated events. There can be no assurance that the initiatives referred to above will be successful.
Contact Information
Hande Asik
Group Director of Communications
pr@veon.com
FAQ
What is the size of VEON's new Kazakh language model (Kaz-LLM)?
Which languages does VEON's Kaz-LLM support?
How many downloads has VEON's previous Kaz-RoBERTA model achieved?
Who are the main partners in VEON's Kaz-LLM development?
Where is VEON's Kaz-LLM model available for developers?
