Seunghyuk Oh
I am an AI Research Engineer at FuriosaAI, where I lead the development of our in-house LLM, from pre-training to post-training. Previously, I received my M.S. from KAIST ALIN Lab, advised by Jinwoo Shin.
Highlights
- AsyncOPD: How Stale Can On-Policy Distillation Be?
- EfficientRollout: System-Aware Self-Speculative Decoding for RL Rollouts
-
ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs
ICLR 2026 project
-
ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification
ICML 2025
Publications
* Equal contribution
2026
-
AsyncOPD: How Stale Can On-Policy Distillation Be?
Wonjun Kang*, Kevin Galim*, Seunghyuk Oh, Minjun Kang, Sanghyun Park, Donghoon Kim, Minjae Lee, Minseo Kim, Rishabh Tiwari, Yuchen Zeng, Hyung Il Koo, Kangwook Lee
-
EfficientRollout: System-Aware Self-Speculative Decoding for RL Rollouts
Minseo Kim*, Minjae Lee*, Seunghyuk Oh, Kevin Galim, Donghoon Kim, Coleman Hooper, Harman Singh, Amir Gholami, Hyung Il Koo, Wonjun Kang
-
ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs
Wonjun Kang*, Kevin Galim*, Seunghyuk Oh*, Minjae Lee, Yuchen Zeng, Shuibai Zhang, Coleman Hooper, Yuezhou Hu, Hyung Il Koo, Nam Ik Cho, Kangwook Lee
-
CATS: Inference-aligned SFT for Diffusion LLMs via Context-sensitivity Aware Trajectory Sampling
ICLR 2026 DeLTa WorkshopfuriosaSeunghyuk Oh, Minjae Lee, Kevin Galim, Minseo Kim, Hyung Il Koo, Wonjun Kang, Hanbaek Lyu, Kangwook Lee
-
TABED: Test-Time Adaptive Ensemble Drafting for Robust Speculative Decoding in LVLMs
EACL 2026 FindingsfuriosaMinjae Lee*, Wonjun Kang*, Byeongkeun Ahn, Christian Classen, Kevin Galim, Seunghyuk Oh, Minghao Yan, Hyung Il Koo, Kangwook Lee
-
UNCAGE: Contrastive Attention Guidance for Masked Generative Transformers in Text-to-Image Generation
IEEE Access 2026
2025
-
Mamba Drafters for Speculative Decoding
EMNLP 2025 Findings
-
ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification
ICML 2025
-
Sparsified State-Space Models are Efficient Highway Networks
TMLR 2025