SKT, strengthens cooperation with NVIDIA…reveals collaboration case for proprietary foundation model

SKT, strengthens cooperation with NVIDIA…reveals collaboration case for proprietary foundation model

SK Telecom [017670] announced on the 22nd that it has revealed cases of cooperation with NVIDIA in AI model development.

SKT conveyed its AI collaboration plans, including the development of A.X K2, at the NVIDIA Nemotron Developer Days held on the 21st.

Both companies have been exchanging across data, infrastructure, and learning environments, starting from the establishment of the A100 GPU-based supercomputer Titan, which SKT unveiled in 2021.

In particular, SKT adopted the NVIDIA Nemotron dataset for model training while carrying out the government's proprietary AI foundation model project last year.

In this process, both companies utilized Megatron LM, a large-scale model distributed learning framework, and NVIDIA NeMo Curator to enhance the stability of AI training.

Both companies shared their experience in creating hyper-scale models with new architectures like MoE (Mixture of Experts) and related development infrastructure information that supports them, establishing a foundation for stable model training.

SKT plans to continue its collaboration with NVIDIA and utilize NVIDIA solutions for the training of A.X K2, which is currently under development.