Kevin Ko
Research Scientist at kakao
Seoul, Korea, Republic of
Summary
🤩
RockstarKevin Ko is an ML-focused research scientist based in Seoul with 7 years of experience building and optimizing large-scale deep learning systems. He contributes to high-impact open-source projects—such as DeepSpeed, Megatron-LM, and Pororo—working on distributed training optimizations, fused kernels, and NLP model reliability for production and research use. At Kakao, he bridges research and engineering, implementing practical NER and dialogue components for Korean-language chatbots and improving model performance across hardware configurations. His contributions include subtle but important fixes (e.g., non-contiguous tensor handling, activation checkpointing, and fused softmax stability) that enable scalable training of massive transformer models. Comfortable across backend systems and model engineering, he combines low-level performance debugging with applied NLP development.
7 years of coding experience
Github Skills (30)
transformers10
named-entity-extraction10
pytorch10
named-entity-recognition10
entity-extraction10
parallelization10
python10
text-summarization10
machine-learning10
inference10
large-language-models10
pipelining10
ml10
deeplearning-ai10
deep-learning10
Programming languages (7)
Github contributions (5)
hyunwoongko/kochat
Jun 2019 - Jun 2021
Opensource Korean chatbot framework
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:1 release, 145 commits, 3 PRs in 1 year 11 months
Contributions summary:Kevin appears to be primarily focused on implementing a Named Entity Recognition (NER) model within the Korean chatbot framework. They have been working on developing LSTM-CRF models, along with associated data processing and model building steps, and improving scenario-specific capabilities. Furthermore, the user has contributed to various scenario implementations (restaurant, translate, weather, wiki), refining the chatbot's ability to process and respond to user inputs.
nlpweb-crawlerdeep-learningsequance-taggingkorean-chatbot
deepspeedai/DeepSpeed
Jun 2021 - Nov 2022
DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.
Role in this project:ML Engineer
ML EngineerContributions:54 reviews, 6 commits, 10 PRs in 1 year 5 months
Contributions summary:Kevin contributed to the DeepSpeed library, focusing on improvements and new features related to deep learning model optimization. Their work involved fixing bugs related to non-contiguous tensors, updating sparse attention operations, and adding flexibility to pipeline parallel modules and the engine. They also addressed bugs related to activation checkpointing and added a new feature for scaling attention based on the inverse layer index, which is crucial for large-scale model training.
billion-parametersfinetuningtrainingmixture-of-expertszero
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.