AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Language Model Networks: Supervision-Efficient Learning through Dense Communication

Shiguang Wu, Yaqing Wang, Quanming Yao

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 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
use LMNet dense communication instead of prompting the same modelQwen2.5-0.5B · General benchmarks averagerelative improvement: +30.5%Improved
SFT on the same training data as LMNet (ablation)Qwen2.5-0.5B · General benchmarks averagerelative improvement: -5.7%Hurt
add Self-Consistency with 16 samples (ablation)Qwen2.5-0.5B · General benchmarks averagerelative improvement: +3.4%No clear effect
add Self-Refine with 16 iterations (ablation)Qwen2.5-0.5B · General benchmarks averagerelative improvement: +0.5%No clear effect
LMNet instead of promptingQwen2.5-0.5B · MATHaccuracy: 19.5 -> 38.8Improved
LMNet instead of promptingQwen2.5-0.5B · GPQAaccuracy: 24.8 -> 25.6No clear effect
replace SFT with LMNetQwen2.5-1.5B · MMLU limited dataDelta Acc: 35.02 -> 37.28Improved
use LMNet instead of latent reasoning COCONUTQwen2.5-1.5B · GSM8K limited dataaccuracy: 48.63 -> 72.67Improved
COCONUT latent reasoning instead of plain promptingQwen2.5-0.5B · MMLU limited dataDelta Acc: 22.50 -> 19.22Hurt
increase vertex model size to 3B and 7B (ablation)LMNet · MMLU and GSM8Kaccuracy: MMLU 7B: 74.2 -> 76.0 vs PromptImproved
vary LMNet depth/width (ablation)LMNet GPT2-M · E2EBLEU: 1/1: 69.1, 1/2/1: 70.5, 1/4/4/1: 68.7No clear effect
LMNet edge-only tuning instead of LoRAGPT2-M · E2EBLEU: 68.9 -> 70.5Improved

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