Seoul National University VLSI Lab · 5 records

dateidtitleauthorsvenue↑
2026-10-152610.19026Joint Subspace Sparse Attention for Efficient Long-Context DecodingJiwon Song et al.arXiv13
2026-09-292609.23033WaveFront Decoding: Parallelized Self-Speculative Decoding for Looped Language ModelsHyeongju Ha et al.arXiv9
2026-09-292609.34060KVCMAS: Efficient KV cache Correction for Shared Context in Multi-Agent SystemsHyesung Jeon et al.arXiv7
2026-09-292609.34054PReCache: Efficient KV Cache Sharing for Multi-LoRA Agents via Low-Rank Precomputation and Neutral ReconstructionHyesung Jeon et al.arXiv7
2026-05-192605.16839CompactAttention: Accelerating Chunked Prefill with Block-Union KV SelectionJiwon Song et al.arXiv12