2 papers
cs.CL2026
Behavior-Preserving KV Cache Compression
Doo Hwan Hwang, Junyoung Jang, Junho Na +2
KV caches are a major bottleneck in long-context inference and long-form generation with large language models. Existing training-free eviction policies largely rely on proxy impor…
cs.LG2026
PS-PPO: Prefix-Sampling PPO for Critic-Free RLHF
Doo Hwan Hwang, Kee-Eung Kim
Reinforcement Learning from Human Feedback (RLHF) for Large Language Models increasingly relies on critic-free methods as a practical alternative to actor--critic training. Despite…