1
SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference
将激活稀疏性与FP4量化巧妙结合,大幅提升LLM推理效率,硬核优化方案来袭!
arXiv:2606.26587v1 Announce Type: cross Abstract: Low-bit floating-point formats and semi-structured sparsity are increasingly supported by modern acc…