Reuse, Don't Recompute: Efficient Large Reasoning Model Inference via Memory Orchestration

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Patel, Daivik, Patel, Shrenik
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914363803172864
author Patel, Daivik
Patel, Shrenik
author_facet Patel, Daivik
Patel, Shrenik
contents Large reasoning models (LRMs) achieve strong accuracy through test-time scaling, generating longer chains of thought or sampling multiple solutions, but at steep costs in tokens and latency. We argue that memory is a core ingredient for efficient reasoning: when evidence already exists, models should think less by reusing structured memory instead of recomputing derivations. We present ENGRAM-R, an inference-time memory layer that integrates typed retrieval with compact fact card representations and explicit citation control. On the LoCoMo benchmark, ENGRAM-R reduces input tokens by 85% and reasoning tokens by 75% compared to full context while maintaining high accuracy. On a multi-hop slice of the LongMemEval benchmark, it achieves similar efficiency with substantial accuracy gains. These results show that memory is not only critical for long-horizon correctness but also a practical lever for efficient reasoning under tight compute, memory, and latency budgets.
format Preprint
id arxiv_https___arxiv_org_abs_2511_12987
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reuse, Don't Recompute: Efficient Large Reasoning Model Inference via Memory Orchestration
Patel, Daivik
Patel, Shrenik
Multiagent Systems
Large reasoning models (LRMs) achieve strong accuracy through test-time scaling, generating longer chains of thought or sampling multiple solutions, but at steep costs in tokens and latency. We argue that memory is a core ingredient for efficient reasoning: when evidence already exists, models should think less by reusing structured memory instead of recomputing derivations. We present ENGRAM-R, an inference-time memory layer that integrates typed retrieval with compact fact card representations and explicit citation control. On the LoCoMo benchmark, ENGRAM-R reduces input tokens by 85% and reasoning tokens by 75% compared to full context while maintaining high accuracy. On a multi-hop slice of the LongMemEval benchmark, it achieves similar efficiency with substantial accuracy gains. These results show that memory is not only critical for long-horizon correctness but also a practical lever for efficient reasoning under tight compute, memory, and latency budgets.
title Reuse, Don't Recompute: Efficient Large Reasoning Model Inference via Memory Orchestration
topic Multiagent Systems
url https://arxiv.org/abs/2511.12987