Sunghun Kim, also known professionally as Sung Kim, is Upstage’s co-founder and chief executive. He combines a background in software-engineering research with AI education and enterprise model development.[1][2][3]
Start with DeepLearningZeroToAll →
Career
Kim earned an electrical-engineering bachelor’s degree at Daegu University in 1996 and a computer-science doctorate at UC Santa Cruz in 2006. He later worked as a postdoctoral researcher at MIT and taught at HKUST. Before Upstage, he led Naver’s Clova AI organization. Upstage was founded in 2020 to help businesses apply AI to their own work.[3][2]
Contributions
Teach the machinery behind deep learning
Kim’s DeepLearningZeroToAll repository accompanies his educational work with runnable examples. It progresses through linear regression, classification, neural networks and convolutional models, with implementations across several frameworks. The repository is useful for learning the operations behind model training, although some examples depend on older library versions.[4]
Scale a language model through depth
Kim co-authored the SOLAR 10.7B paper. It describes depth up-scaling: increasing model depth and then continuing pretraining, without introducing mixture-of-experts routing. The team also develops an instruction-tuned version and reports comparisons on language tasks. The method offers a concrete engineering route for adapting an existing model; the paper’s results do not establish superiority over later models.[5]
Tools, writing & talks
Tools
Start here
DeepLearningZeroToAll
github.comTutorial code for the fundamentals of training neural networks.
Start with linear regression and follow the examples in order. Basic Python and algebra are enough; use a suitable historical environment or adapt the examples to current libraries.
Educator and repository owner[4]
Writing, research & talks
- SOLAR 10.7Barxiv.org
A model-scaling paper explaining depth up-scaling and continued pretraining.
Co-author[5]