12 papers
AsyncOPD: How Stale Can On-Policy Distillation Be?
Wonjun Kang, Kevin Galim, Seunghyuk Oh +9
On-policy distillation (OPD) trains a student on its own rollouts guided by teacher feedback and is becoming increasingly important for large language model (LLM) post-training. Li…
ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs
Wonjun Kang, Kevin Galim, Seunghyuk Oh +8
While most autoregressive LLMs are constrained to one-by-one decoding, diffusion LLMs (dLLMs) have attracted growing interest for their potential to dramatically accelerate inferen…
EfficientRollout: System-Aware Self-Speculative Decoding for RL Rollouts
Minseo Kim, Minjae Lee, Seunghyuk Oh +7
Reinforcement learning (RL) has become a representative post-training paradigm for LLMs, enabling strong reasoning and agentic capabilities. However, rollout generation remains a d…
LoSA: Locality Aware Sparse Attention for Block-Wise Diffusion Language Models
Haocheng Xi, Harman Singh, Yuezhou Hu +9
Block-wise diffusion language models (DLMs) generate multiple tokens in any order, offering a promising alternative to the autoregressive decoding pipeline. However, they still rem…
Draft-based Approximate Inference for LLMs
Kevin Galim, Ethan Ewer, Wonjun Kang +3
Optimizing inference for long-context large language models (LLMs) is increasingly important due to the quadratic compute and linear memory cost of Transformers. Existing approxima…
TABED: Test-Time Adaptive Ensemble Drafting for Robust Speculative Decoding in LVLMs
Minjae Lee, Wonjun Kang, Byeongkeun Ahn +6
Speculative decoding (SD) has proven effective for accelerating LLM inference by quickly generating draft tokens and verifying them in parallel. However, SD remains largely unexplo…