SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance 文章

ArXiv CS.AI2026-06-09NEWSen作者: Rya Sanovar, Srikant Bharadwaj, Hritvik Taneja, Moinuddin Qureshi
查看原文 →
关系图谱

SIFT: Selective-Index For Fast Compute of RAG Prefill by Exploiting Attention Invariance · 相关技术

相关技术

GPUagentic retrieval-augmented generationLLMAttention InvarianceKV tensortime to first tokenSIFT