Research map / Large language models / Retrieval and knowledge
Pre-training Limited Memory Language Models with Internal and External Knowledge
Linxi Zhao, Sofian Zalouk, Christian Belardi, Justin Lovelace, Jin Peng Zhou, Ryan Thomas Noonan, Dongyoung Go, Kilian Q Weinberger, Yoav Artzi, Jennifer J. Sun
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 |
|---|---|---|---|
| externalize facts to a database with masked-loss lookup pretraining | LLaMA2-382M · FactScore | FactScore: 14.0 -> 31.9 | Improved |
| externalize facts with lookup pretraining | LLaMA2-382M · T-REx | EM: 52.0 -> 58.1 | Improved |
| externalize facts with lookup pretraining | GPT2-124M · PopQA long-tail | accuracy: 18.51 -> 49.89 | Improved |
| LMLM instead of Standard (validation perplexity) | GPT2/LLaMA2 small models · Wikipedia validation | perplexity (Dynamic): average reduction 1.98 | Improved |
| LMLM instead of Standard (NLU) | GPT2/LLaMA2 small models · five NLU tasks | accuracy | No clear effect |
| disable external database at inference (ablation) | LMLM LLaMA2-382M · FactScore | FactScore: 31.9 -> 12.8 | Hurt |
| disable external database at inference (ablation) | LMLM LLaMA2-382M · T-REx | EM: 58.1 -> 38.5 | Hurt |
| LMLM vs RAG with top-4 Wikipedia articles | GPT2-355M · T-REx | EM: 75.8 (RAG) vs 58.7 (LMLM) | Hurt |
| LMLM vs RAG | GPT2-355M · PopQA | accuracy: 37.5 (RAG) -> 52.0 (LMLM) | Improved |
| increase knowledge offloading ratio toward 100% (ablation) | LMLM · perplexity and factual precision | perplexity, factual precision | Improved |
| mask retrieved values from loss, delete forget-set entries (no training) | LMLM LLaMA-3.2-1B · TOFU forget 5% | forget quality / utility | Improved |
Closest papers by meaning
- Memory Grafting: Scaling Language Model Pre-training via Offline Conditional Memory (NeurIPS 2026 · similarity 0.89)
- Mitigating Memorization in Language Models (ICLR 2025 · similarity 0.88)
- Logically Consistent Language Models via Neuro-Symbolic Integration (ICLR 2025 · similarity 0.88)
- Steering Information Utility in Key-Value Memory for Language Model Post-Training (NeurIPS 2025 · similarity 0.88)
- MiniPLM: Knowledge Distillation for Pre-training Language Models (ICLR 2025 · similarity 0.88)
- Once Read is Enough: Domain-specific Pretraining-free Language Models with Cluster-guided Sparse Experts for Long-tail Domain Knowledge (NeurIPS 2024 · similarity 0.88)
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.
