Our new preprint, Tailoring the Quantization Space for 1-Bit KV Cache Compression, is now public. It introduces a 1-bit KV-cache quantization method that works remarkably well. Check it out!
Hello, I’m
Minsoo Cheong
I’m an integrated M.S./Ph.D. student in Computer Science and Engineering at Seoul National University.
My research focuses on efficient machine learning systems. My recent work explores efficient language model inference, particularly KV-cache optimization, diffusion language models, and low-bit quantization. I am advised by Prof. Sungjoo Yoo at CMALab.
Recent news
Recent news
My first first-author paper, EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models, was accepted to NeurIPS 2026 for a poster presentation. 🎉
Research
Efficient machine learning systems.
My research focuses on reducing the compute and memory cost of machine learning systems. I am currently working on efficient language model inference, with a focus on KV-cache optimization and low-bit quantization. My broader interests include LLM serving systems and alternative model architectures, particularly diffusion language models and linear attention.
Selected publications
Selected publications
01
02
Tailoring the Quantization Space for 1-Bit KV Cache Compression
A tailored quantization space for compressing the KV cache to one bit.
03
KV-Rescue: Recovering Reasoning Language Model KV Eviction Loss via Stepwise Interleaving
Stepwise interleaving to recover reasoning-model performance lost through KV-cache eviction.
04
Experience
Systems research, backed by production engineering.
Experience across model inference, backend platforms, and open-source runtimes.
Jun — Aug 2024
NAVER Corporation
Research Engineer Intern LLM Solution Team
Built a Kubeflow pipeline for automated LLM inference benchmarking across hardware, batching, sequence length, quantization, and speculative decoding. Implemented llama.cpp support for an internal language model.
Jun 2022 — Jun 2023
Mathpresso
Backend Software Engineer Content Platform Team
Built Kotlin and Go content-production and distribution systems and gRPC APIs. Contributed to three release cycles of a digital school-exam product that achieved a 60-fold increase in sales.
May 2021 — Jun 2022
Bucketplace
Backend Software Engineer Commerce Service Team
Developed commerce backend services using Kotlin and Spring Boot, including bulk product management and microservice data aggregation.
Open source
Ideas that run outside the paper.
Selected work on LLM inference and Kotlin developer infrastructure.
CollaboratorMar 2024 — Present
llama.cpp
Contributions to stochastic speculative decoding, the retrieval example, and EXAONE model support in the widely used C/C++ LLM inference runtime.
View contributionsOpen source
krotoDC
A protoc plugin for Kotlin data classes and coroutine-friendly gRPC services and stubs.
View on GitHubBackground
Education
Seoul National UniversitySep 2025 — Present
Integrated M.S./Ph.D. in Computer Science and Engineering
Advisor: Prof. Sungjoo Yoo · Computing and Memory Architecture LaboratorySeoul National UniversityMar 2019 — Aug 2025
B.S. in Computer Science and Engineering, Cum Laude
Talks & service
EntropyCache poster presentations at NeurIPS and the ICML AdaptFM Workshop; talk at the Samsung MX Research Project Exchange Seminar.
Reviewer, NeurIPS Workshop on On-Device Intelligence: Foundation Models under Real-World Constraints.