AI Paper Atlas

Research map / Large language models / Agents, code and math

LightWM: Training-Free Hierarchical Working Memory for Small Language Model Agents

Ziyi Wang, Haonan Jin, Zian Wang, Wendong Wang, Lanshan Zhang

ICML 2026 · Large language models › Agents, code and math · approach: Agent instruction-following benchmarks

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
add subtask-specific action templates (ablation)LightWM · ALFWorld valid unseen, Qwen3-4B (FP16)success rate: 0.821 -> 0.910Improved
add rule-constrained patch updates (ablation)LightWM · ALFWorld valid unseen, Qwen3-4B (FP16)success rate: 0.766 -> 0.910Improved
add action templates and patch constraints on top of HWM only (ablation)LightWM · ALFWorld valid unseen, Qwen3-4B (FP16)success rate: 0.697 -> 0.910Improved
add hierarchical goal-conditioned working memory (HWM only vs MemGPT) (ablation)Qwen3-4B (FP16) · ALFWorld valid unseen, Qwen3-4B (FP16)success rate: 0.317 -> 0.697Improved
add rule-constrained patch updates (output tokens per call) (ablation)LightWM · ALFWorld valid unseen, Qwen3-4B (FP16)output tokens per call: 139 -> 32 output tokens/callImproved
use LightWM instead of ReActQwen3-4B (FP16) · ALFWorld valid unseensuccess rate: 0.062 -> 0.910Improved
use LightWM instead of ReActGemini-2.5 Flash · ALFWorld valid unseensuccess rate: 0.786 -> 0.945Improved
use LightWM instead of ReActQwen3-8B (FP16) · ALFWorld valid unseensuccess rate: 0.579 -> 0.828Improved
use Simply Summarize instead of ReActQwen3-8B (FP16) · ALFWorld valid unseensuccess rate: 0.579 -> 0.469Hurt
use summarization-based memory (Simply Summarize) instead of ReActGemini-2.5 Flash · ALFWorld valid unseensuccess rate: 0.786 -> 0.531Hurt
use LightWM instead of MemGPTQwen3-4B (FP16) · ALFWorld valid unseensuccess rate: 0.317 -> 0.910Improved
use FP16 instead of FP8 inferenceLightWM · ALFWorld valid unseen, Qwen3-4Bsuccess rate: 0.821 -> 0.910Improved

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