Research map / Large language models / Language, safety and interpretability
Revisiting Long-context Modeling from Context Denoising Perspective
Zecheng Tang, Baibei Ji, Juntao Li, Lijun Wu, Haijia Gui, Min Zhang
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 |
|---|---|---|---|
| add context denoising training (CDT) on top of SFT for long-context alignment | Llama-3.1-8B-Instruct · LongBench-E | average score: 48.61 -> 50.92 | Improved |
| replace plain SFT with CDT (ablation) | Llama-3.1-8B-Instruct · LongBench-E | average score: 48.72 -> 50.92 | Improved |
| use CDT instead of LongCE for context window scaling (ablation) | Llama-3-8B-Base (8K to 64K) · LongBench-E | average score: 34.62 -> 39.31 | Improved |
| use LongCE or CE objective for post-training on base model (ablation) | Llama-3.1-8B-Base · LongBench-E | average score: 37.13 -> 36.90 (LongCE); 37.13 -> 37.55 (CE) | No clear effect |
| add CDT to post-training of base model | Llama-3.1-8B-Base · LongBench-E | average score: 37.13 -> 38.89 | Improved |
| add CDT | Llama-3.1-8B-Instruct · RULER 128K | average score: 76.71 -> 78.72 | Improved |
| add CDT | Llama-3.1-8B-Instruct · GovReport LongPPL | LongPPL: 4.05 -> 2.36 | Improved |
| add CDT | Llama-3.1-8B-Instruct · GovReport PPL | PPL: 5.52 -> 5.64 | Hurt |
| add CDT | Llama-3.1-8B-Instruct · BABILong avg | accuracy: 40.67 -> 43.30 | Improved |
| add CDT | Llama-3.1-8B-Instruct / Qwen3-8B / Mistral-V0.3 · LongBench-E | average score: Qwen3-8B 38.85 -> 41.11; Mistral 44.06 -> 47.58 | Improved |
| add CDT | Meta-Llama-3.1-8B · HELMET | average score: 46.5 -> 49.7 | Improved |
| use embedding-gradient detection (CDT) instead of attention-based or LongPPL detection of critical tokens (ablation) | CDT critical token detection · synthetic long-form reasoning task | critical token detection | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- YaRN — YaRN: Efficient Context Window Extension of Large Language Models (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Little Goes a Long Way: Efficient Long Context Training and Inference with Partial Contexts (ICLR 2025 · similarity 0.89)
- What is Wrong with Perplexity for Long-context Language Modeling? (ICLR 2025 · similarity 0.88)
- Short-Context Dominance: How Much Local Context Natural Language Actually Needs? (NeurIPS 2026 · similarity 0.88)
- LongMamba: Enhancing Mamba's Long-Context Capabilities via Training-Free Receptive Field Enlargement (ICLR 2025 · similarity 0.87)
- NExtLong: Toward Effective Long-Context Training without Long Documents (ICML 2025 · similarity 0.87)
- A Controllable Examination for Long-Context Language Models (NeurIPS 2025 · similarity 0.87)
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.
