Research map / Large language models / Language, safety and interpretability
Discovering Decoupled Functional Modules in Large Language Models
Yanke Yu, Jin Li, Ying Sun, Ping Li, Zhefeng Wang, Yi Zheng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use IterD instead of K-Means for module discovery | Qwen2.5-7B-Instruct · Infinity-Instruct activations, K=5 | L(F) x1e6: 29.6 -> 64.6 | Improved |
| use IterD instead of Spectral Co-clustering | Qwen2.5-7B-Instruct · K=10 | L(F) x1e6: 16.9 -> 25.2 | Improved |
| use IterD discovered modules to build features for function-label classification | Logistic Regression on Qwen2.5-7B-Instruct · K=5 | accuracy: 0.5121 -> 0.6000 | Improved |
| use IterD instead of baselines for classification features at K=20 | Qwen2.5-1.5B-Instruct, SVC · K=20 | accuracy: 0.7986 -> 0.8236 (K-Means to IterD) | Improved |
| use IterD instead of Mini-Batch K-Means for classification features | Qwen2.5-7B-Instruct · K=20, Mini-Batch K-Means vs IterD | accuracy: 0.8036 -> 0.8214 (SVC) | Improved |
| increase number of modules K | IterD · Qwen2.5-7B-Instruct | L(F): 64.6 (K=5) -> 25.2 (K=10) | Hurt |
| increase number of modules K | IterD · Qwen2.5-7B-Instruct | classification accuracy: 0.6000 (K=5) -> 0.8179 (K=20), LR | Improved |
Closest papers by meaning
- Towards Global-level Mechanistic Interpretability: A Perspective of Modular Circuits of Large Language Models (ICML 2025 · similarity 0.86)
- How do Large Language Models Handle Multilingualism? (NeurIPS 2024 · similarity 0.86)
- Sparsity-Guided Holistic Explanation for LLMs with Interpretable Inference-Time Intervention (AAAI 2024 · similarity 0.85)
- Extracting Interpretable Task-Specific Circuits from Large Language Models for Faster Inference (AAAI 2025 · similarity 0.85)
- Heads up! Large Language Models Can Perform Tasks Without Your Instruction via Selective Attention Head Masking (ICML 2025 · similarity 0.85)
- ToolACE: Winning the Points of LLM Function Calling (ICLR 2025 · similarity 0.85)
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.
