Junlin Mu

Online Pseudo-average Shifting Attention(PASA) for Robust Low-precision LLM Inference: Algorithms, Numerical Analysis and Performance

SALS: Sparse Attention in Latent Space for KV Cache Compression