Research map / Large language models / Language, safety and interpretability
L-CUBE: Isolating Long-Context Capacity from Knowledge with Controllable Mutual Information Scaling
Zhuo Chen, Oriol Mayné i Comas, Zhuotao Jin, Di Luo, Marin Soljacic
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 full-attention transformer instead of state space model | long-context utilization on L-CUBE · L-CUBE maximum utilized context length | maximum utilized context length: 3,072 -> 12,288 | Improved |
| scale up SSM model size (Mamba2 785M, Mamba 793M) (ablation) | Mamba and Mamba2 · L-CUBE long-context utilization | long-context utilization | Improved |
| use Mamba2 instead of Mamba at larger size | SSMs 785M vs 793M · L-CUBE long-context utilization | long-context utilization | Improved |
| use GPT-2 learned positional embeddings instead of GPT-NeoX rotary embeddings at small scale | GPT-2 64M vs GPT-NeoX 70M · L-CUBE long-context utilization | long-context utilization | Improved |
| use log-linear attention instead of SSMs | LLA 176M · L-CUBE long-context utilization | long-context utilization | Mixed |
| exceed sliding window size in windowed attention (ablation) | Mistral window 64/256/4096 · L-CUBE long-context utilization | long-context utilization | Hurt |
| augment pretraining with L-CUBE synthetic sequences (ablation) | Pythia-410M on PG19 (1% long real data) · PG19 long-sequence loss at 4096 tokens | loss | Improved |
Closest papers by meaning
- L$^2$M: Mutual Information Scaling Law for Long-Context Language Modeling (NeurIPS 2025 · similarity 0.92)
- Data Engineering for Scaling Language Models to 128K Context (ICML 2024 · similarity 0.89)
- A Little Goes a Long Way: Efficient Long Context Training and Inference with Partial Contexts (ICLR 2025 · similarity 0.89)
- A Controllable Examination for Long-Context Language Models (NeurIPS 2025 · similarity 0.89)
- What is Wrong with Perplexity for Long-context Language Modeling? (ICLR 2025 · similarity 0.88)
- Longer Context, Deeper Thinking: Uncovering the Role of Long-Context Ability in Reasoning (NeurIPS 2025 · 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.
