SWAN: Sparse Winnowed Attention for Reduced Inference Memory via Decompression-Free KV-Cache Compression

Documenti analoghi

• AQUA: Attention via QUery mAgnitudes for Memory and Compute Efficient Inference in LLMs
  di: S, Santhosh G, et al.
  Pubblicazione: (2025)
• ManifoldKV: Training-Free KV Cache Compression via Euclidean Outlier Detection
  di: Datta, Debajyoti, et al.
  Pubblicazione: (2026)
• FreeKV: Boosting KV Cache Retrieval for Efficient LLM Inference
  di: Liu, Guangda, et al.
  Pubblicazione: (2025)
• Eigen Attention: Attention in Low-Rank Space for KV Cache Compression
  di: Saxena, Utkarsh, et al.
  Pubblicazione: (2024)
• HCAttention: Extreme KV Cache Compression via Heterogeneous Attention Computing for LLMs
  di: Yang, Dongquan, et al.
  Pubblicazione: (2025)