Sungwon Kim

Hello! I am Sungwon Kim, and I am currently a research assistant at Johns Hopkins University. My current research focuses on KV cache efficiency for large language models: eviction, compression, and sparse attention. I’m also open to exploring other directions.

I have graduated with a BS in Computer Science and Applied Mathematics and Statistics from Johns Hopkins in 2024, and a Masters in Computer Science from Johns Hopkins in 2025.

news

Oct 19, 2026	What Survives Eviction? Information Recall Locality in Transformer KV Caches has been accepted to TMLR.
Jul 08, 2026	BiomedSQL: Text-to-SQL for Scientific Reasoning on Biomedical Knowledge Bases has been accepted to COLM 2026.
Jan 31, 2026	CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research has been accepted to The Lancet Digital Health.
Aug 20, 2025	Challenging the Evaluator: LLM Sycophancy Under User Rebuttal was accepted to EMNLP Findings.
May 20, 2025	Graduated with Masters in Computer Science from Johns Hopkins University.

latest posts

Nov 21, 2025	Old NLP Paper Reviews
Nov 07, 2025	Showcase of EMNLP Sycophancy Papers
Oct 22, 2025	Torch MPS Basic Speedup Test

selected publications

TMLR

What Survives Eviction? Information Recall Locality in Transformer KV Caches

Sungwon Kim and Daniel Khashabi

Transactions on Machine Learning Research, 2026

Bib HTML

@article{kim2026what,
  title = {What Survives Eviction? Information Recall Locality in Transformer {KV} Caches},
  author = {Kim, Sungwon and Khashabi, Daniel},
  journal = {Transactions on Machine Learning Research},
  issn = {2835-8856},
  year = {2026},
  url = {https://openreview.net/forum?id=eAMLpW74rV},
}

Lancet Digital Health

CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research

Owen Bianchi, Maya Willey, Chelsea X. Alvarado, and 21 more authors

Lancet Digital Health, Jan 2026

DOI Bib HTML

@article{bianchi2025cardbiomedbench,
  title = {CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research},
  author = {Bianchi, Owen and Willey, Maya and Alvarado, Chelsea X. and Danek, Ben and Khani, Mahsa and Kuznetsov, Nicole and Dadu, Aanya and Shah, Saniya and Koretsky, Mathew J. and Makarious, Mary B. and Weller, Claire and Levine, Kelsey S. and Kim, Sungwon and Jarreau, Paul and Vitale, Domenico and Marsan, Emilie and Iwaki, Hiroto and Leonard, Heather and Bandres-Ciga, Sara and Singleton, Andrew B. and Nalls, Mike A. and Mokhtari, Soheil and Khashabi, Daniel and Faghri, Faraz},
  journal = {Lancet Digital Health},
  year = {2026},
  month = jan,
  doi = {10.1016/j.landig.2025.100943 },
  url = {https://www.sciencedirect.com/science/article/pii/S2589750025001256},
}

COLM

BiomedSQL: Text-to-SQL for Scientific Reasoning on Biomedical Knowledge Bases

Mathew J. Koretsky, Maya Willey, Owen Bianchi, and 7 more authors

In Third Conference on Language Modeling, 2026

arXiv Bib HTML

@inproceedings{koretsky2026biomedsql,
  title = {Biomed{SQL}: Text-to-{SQL} for Scientific Reasoning on Biomedical Knowledge Bases},
  author = {Koretsky, Mathew J. and Willey, Maya and Bianchi, Owen and Alvarado, Chelsea X and Nayak, Tanay and Kuznetsov, Nicole and Kim, Sungwon and Nalls, Michael and Khashabi, Daniel and Faghri, Faraz},
  booktitle = {Third Conference on Language Modeling},
  year = {2026},
  url = {https://openreview.net/forum?id=SC7liryuqr},
}

EMNLP

Challenging the Evaluator: LLM Sycophancy Under User Rebuttal

Sungwon Kim and Daniel Khashabi

In Findings of the Association for Computational Linguistics: EMNLP 2025, Nov 2025

arXiv Bib HTML

@inproceedings{kim2025challenging,
  title = {Challenging the Evaluator: LLM Sycophancy Under User Rebuttal},
  author = {Kim, Sungwon and Khashabi, Daniel},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2025},
  year = {2025},
  month = nov,
  address = {Suzhou, China},
  publisher = {Association for Computational Linguistics},
  pages = {22461--22478},
  isbn = {979-8-89176-335-7},
  url = {https://aclanthology.org/2025.findings-emnlp.1222/},
}