A.X K2 Technical Report
Published in SK Telecom Tech Report, 2026
Overview
During my research internship at SK Telecom (March 2026 – June 2026), I participated in the post-training process for SKT’s proprietary foundation model, A.X K2.
This technical report details the development, architecture, training strategies, alignment methodologies, and benchmark performance of the A.X K2 model.
Key Contributions
- Post-Training Alignment: Contributed to the post-training process for large-scale AI models.
- Multimodal Reasoning Pipeline: Designed dataset construction strategies and data generation pipelines to enhance multimodal reasoning capabilities.
- Large-Scale Data Scaling: Constructed Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) datasets at a scale of 1M+ samples.
Resources
- 📄 Technical Report PDF: A.X K2 Technical Report
- 🐙 GitHub Repository: SKT-AI/A.X-K2