Yandex Develops and Open-Sources an LLM Training Tool That Saves up to 20 Per cent of GPU Resources

Mumbai (Maharashtra) [India], June 12: Yandex, a global tech leader, proudly announces the release of YaFSDP, an open-source tool designed to revolutionize the large language model training in India. This innovative tool significantly accelerates training times and reduces hardware consumption, making it highly relevant for India’s rapidly growing AI and tech sectors.

Unleashing the Power of YaFSDP

Yandex’s YaFSDP eliminates GPU communication inefficiencies, ensuring that training requires only necessary processor memory and making GPU interactions uninterrupted. YaFSDP is currently the most effective publicly available tool for enhancing GPU communication and reducing memory usage in LLM training, offering a speedup of up to 26 per cent compared to FSDP, depending on the architecture and number of parameters. Reducing the training time for LLMs through the use of YaFSDP can result in savings of up to 20 per cent in GPU resources, which can lead to savings of over 84,000,000 Indian Rupees monthly.

Key Benefits for India:

• Enhanced Efficiency: YaFSDP optimizes network usage and reduces memory load, ensuring faster and more efficient training of AI models. This is crucial for India’s AI-driven industries such as finance, healthcare, and education.
• Cost Savings: By reducing computational resource requirements, YaFSDP lowers the cost of AI training, making advanced AI technologies more accessible.
• Environmental Impact: With reduced energy consumption, YaFSDP contributes to a smaller carbon footprint, aligning with India’s sustainability goals. Empowering Innovators “YaFSDP has the potential to drive significant advancements in the tech landscape,” said Mikhail Khruschev, a senior developer at Yandex and part of the team behind YaFSDP. “By improving training efficiency, we aim to empower developers, researchers, and companies to build more sophisticated and powerful AI models.” Relevance to the Indian Market:

• Transforming Industries: YaFSDP can revolutionize AI applications in key Indian sectors. For instance, in healthcare, faster AI training can lead to more accurate diagnostic tools; in finance, it can enhance fraud detection systems.

• Supporting Startups: The cost savings and efficiency gains from YaFSDP are particularly beneficial for Indian startups, enabling them to compete on a global scale without the burden of high computational costs.

• Academic Collaboration: Indian academic institutions can leverage YaFSDP to advance their AI research, fostering innovation and producing cutting-edge research.

• Success Stories and Applications YaFSDP has shown impressive results in optimizing training for models like Llama 2 and Llama 3, demonstrating significant speedup and efficiency improvements. In a pre-training scenario with a 70 billion parameter model, YaFSDP saved resources equivalent to 150 GPUs, highlighting its potential for large-scale AI projects.

• Commitment to Community Yandex is a global technology company that builds intelligent products and services powered by machine learning. Yandex is committed to supporting the Indian AI community by providing comprehensive documentation, tutorials, and a vibrant community forum for YaFSDP users. The open-source nature of YaFSDP encourages collaboration, enabling Indian developers to contribute to and benefit from global advancements in AI technology.

• Since 1997, Yandex has been delivering world-class, locally relevant search and information services and has also developed market-leading on-demand transportation services, navigation products, and other mobile applications for millions of consumers across the globe. The company’s goal is to help consumers and businesses better navigate the online and offline world. Accessing YaFSDP YaFSDP is freely available on Github . Indian developers and researchers are encouraged to explore its capabilities and join the community of innovators transforming the future of AI.