AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Understand and Accelerate Memory Processing Pipeline for Large Language Model Inference

Zifan He, Rui Ma, Yizhou Sun, Jason Cong

ICML 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
offload memory-processing steps (relevancy, retrieval) of sparse attention to FPGAGPU-FPGA system · sparse attention decoding, up to 1M tokensend-to-end speedup: 1.04-1.49xImproved
offload BM25 and top-k of single-stage RAG to FPGAGPU-FPGA system · DRAGIN/FLARE/FS-RAGmemory processing speedup: 5.1-6.6x over BM25SImproved
offload retrieval of two-stage RAG to FPGA (reranker stays on GPU)GPU-FPGA system · two-stage RAGmemory processing speedup: 1.1-2.1xMixed
offload MemAgent decoding to FPGA in prefill-decode disaggregationGPU-FPGA system · MemAgent, batch size 1speedup: 1.8xImproved
increase batch size from 1 to 32 for sparse attention and RAG (ablation)GPU-FPGA system · LServe geomean speedupspeedup: 1.19x -> 1.83xImproved
increase batch size from 1 to 32 for MemAgent (ablation)GPU-FPGA system · MemAgent geomean speedupspeedup: 1.85x -> 0.13xHurt
increase batch size from 1 to 32 for Memory as Context (ablation)GPU-FPGA system · HMT/Titans geomean speedupspeedup: 1.48x -> 1.15xHurt
sequence length beyond 1M tokens on FPGA (ablation)GPU-FPGA system · LServe and DeepSeek AttentionspeedHurt
use threshold instead of top-k selection in SeerAttention-R on FPGA (ablation)GPU-FPGA system · SeerAttention-R kernel speedupkernel speedup: 1.8-2.2x -> 2.6-4.9xImproved
run GPU-FPGA instead of GPU only for MemAgent energyGPU-FPGA system · MemAgentenergy per request: 13662 -> 3202 J/reqImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy