多智能体 LLM 推理中的持久记忆：代价、收益与可辨识性

View PDF HTML (experimental)

Abstract:Decomposing long-context inference across cooperating agents bounds the active KV cache per call rather than total evidence, which matters when KV-cache memory binds. Many such systems add a persistent tier storing and recalling reasoning traces, usually validated by an ablation reporting an accuracy gain. We measure both on one three-tier agent architecture. Decomposition delivers: peak KV working set of 14.3 MiB per query against 35.5 and 35.3 MiB for single-pass and retrieval-augmented baselines. The persistent tier does not: across eight controlled dataset pairs at n=100 per arm it costs +0.368 MiB [+0.167, +0.590] of peak cache and produces no detectable accuracy change (+0.015, 95% CI [-0.011, +0.046]). We argue the null is structural: single-question benchmarks supply each item with its own evidence and score it independently, and correctness requires resetting stored traces between conditions, so recall has nothing informative to retrieve. Reaching it took four measurement corrections -- three inflating the apparent benefit, the fourth making an effect that size look resolvable -- none visible in the results table. We give the conditions an agent-memory ablation must satisfy and detection procedures that need no knowledge of the specific defect.

Comments:	13 pages, 1 figure. Accepted as a poster at the Machine Learning for Systems Workshop, NeurIPS 2026
Subjects:	Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
ACM classes:	C.4; I.2.7; I.2.11
Cite as:	arXiv:2610.07782 [cs.AI]
	(or arXiv:2610.07782v1 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2610.07782 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Hochan Son [view email]
[v1] Tue, 6 Oct 2026 05:17:44 UTC (1,598 KB)