Where is Vietnam in the fierce global AI race?

According to a report from Grand View Research, the global large language model market size is estimated to reach USD 5,617 million by 2024 and is expected to grow at a compound annual growth rate (CAGR) of 36.9% from 2025 to 2030. Meanwhile, the world is also increasingly interested in AI with much attention focused on countries with many leading AI model developers such as the US and China.

Southeast Asia has received less attention, but is also making strong progress. Despite limited data resources compared to popular global languages such as English and Chinese, companies in the region have begun to seize the opportunity to build their own LLMs. Specifically, Singapore has launched its first model, the SEA-LION LLM, with 13% of its data set trained in Southeast Asian languages.

In Vietnam, Zalo has built a domestic AI model and launched it for the first time in 2023. With the goal of being self-sufficient in AI technology and reducing dependence on foreign technology, the model with 13 billion parameters was developed by Zalo using a training technique from scratch - deploying all processes from parameter initialization, deciding on model architecture to training algorithms on a certain data set.

This training technique helps Vietnamese people master and control the entire training process as well as the model. Up to now, the birth of this model has made Vietnam one of the few countries in Southeast Asia to own a domestic LLM.

Not possessing the potential of the world's largest technology companies, the early stages of developing a large language model in Vietnam encountered many difficulties. In addition to the problem of data shortage because Vietnamese is also classified as having dozens of times poorer data resources than English or Chinese, Vietnamese engineers are also not fully equipped with the necessary server infrastructure.

Meanwhile, major companies in the world have owned thousands of the latest GPUs from Nvidia. In particular, Vietnam also has limitations in human resources and experience in training LLM models when compared to developed countries in the world.

However, Zalo still has a suitable development strategy to help overcome difficulties step by step. Thanks to the computing infrastructure equipped with 8 DGX H100 servers, Zalo's LLM model was developed directly using Nvidia's newest and most scarce GPU line at that time with a performance of up to 256 petaFLOPS (FLoating-point Operations Per Second - one petaFLOP is equivalent to 10 million billion calculations/second).

Zalo also increased the development of Vietnamese data to compensate for the market shortage, while equipping Vietnamese engineers with knowledge and LLM training capacity through a series of studies conducted on small civil GPUs. This creates a foundation ready for when owning a larger computing infrastructure later.

The right development orientation has helped Zalo successfully develop a large language model with 7 billion parameters focusing on Vietnamese after only 6 months of training in 2023, achieving 150% capacity compared to OpenAI's GPT3.5 according to VMLU's assessment - Vietnamese Multitask Language Understanding Benchmark Suite for Large Language Models (LLM) Vietnamese Language Proficiency Assessment and Ranking Platform.

By the end of 2024, Zalo's large language model had risen to the number 2 position on the VMLU's top-ranked models. Specifically, the model ranked only behind Meta's Llama-3-70B, officially surpassing big names such as GPT-4 (OpenAI), gemma-2-9b-it (Google), microsoft/Phi-3-small-128k-instruct (Microsoft).

“Despite having a difficult starting point compared to other large companies in the world, Zalo still decided to enter the game early with the goal of successfully developing Vietnam’s own AI model. We have consulted with researchers and engineers at many leading research institutes in the world to have a suitable development strategy. The successful milestones to date are the motivation for Zalo engineers to continue optimizing the model to be larger in quantity and better in quality,” said Dr. Nguyen Truong Son, Director of Science at Zalo AI.

The successful milestones in researching and training the model have motivated Zalo engineers to apply the LLM model to develop useful applications for Vietnamese users. A typical example is the Kiki Info Q&A Assistant, which is being operated as an OA - Official Account on Zalo. The Assistant provides 3 features including Q&A on various topics such as science, history, traffic laws, etc.; Content creation such as writing, composing emails, posting on social networks, etc.; and Entertainment by providing suggestions on tourist destinations, music, books, etc.

According to Zalo statistics, Kiki Info assistant has had 1 million users accessing OA accounts on Zalo in less than 2 months.

Another application of this large language model is AI cards, which are commonly used by users to send greetings to relatives and friends on important occasions, and have now reached the milestone of 15 million cards created and sent in just 2 months.

The applications are highly appreciated by users for their smart and convenient experiences, helping to shorten search and decision-making time, as well as enhancing the quality of connection with relationships in life.

Towards a new era - the era of national growth, the Party and Government are determined to shape and promote the era of technological development of the country, demonstrated through Resolution 57-NQ/TW on Breakthrough in Science, Technology, Innovation and National Digital Transformation.

In an effort to contribute to the development of the country's technology, Zalo not only aims to master advanced technology in general and AI in particular, but also accompanies the Vietnamese technology community to promote the creation of AI technology products by Vietnamese people to serve Vietnamese people.

Specifically, with the “AI-First” strategy, Zalo is currently the pioneering organization in mastering Artificial Intelligence technology in Vietnam. From 2022 to present, Zalo has developed a series of practical applications from AI to help upgrade user experience such as Text-to-Voice, Text-to-Speech, Voice-to-Text, Zalo AI Avatar, zSticker AI…

Sharing his belief in the future of Vietnamese technology, Zalo Founder Vuong Quang Khai affirmed: “The beauty of Artificial Intelligence also comes from the positive impacts on people's daily lives. We are here to share our successes, aspirations and inspire new ways to develop life-changing technology.”

The founder of Zalo also affirmed: "In terms of AI application development, I am quite optimistic and believe that Vietnamese people are not inferior to anyone, not only making domestic applications but also exporting them to the world."

Source: https://znews.vn/viet-nam-o-dau-giua-cuoc-dua-ai-khoc-liet-tren-toan-cau-post1563045.html