Research map / Large language models / Language, safety and interpretability
LiteLong: Resource-Efficient Long-Context Data Synthesis for LLMs
Junlong Jia, Xing Wu, Chaochen Gao, Ziyang Chen, Zijia Lin, Zhongzhi Li, Weinong Wang, Haotian Xu, Donghui Jin, Debing Zhang, Binghui Guo
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 BISAC categories instead of GPT-4o generated category system (ablation) | LiteLong · HELMET + RULER | average score: 59.94 -> 61.90 | Improved |
| add multi-agent debate for topic generation (ablation) | LiteLong · HELMET + RULER at 128K | average score: 56.33 -> 58.59 | Improved |
| increase topic count beyond 0.5x (ablation) | LiteLong · HELMET + RULER, 2B tokens | average score: 49.01 (0.5x) -> 48.56 (1x) | No clear effect |
| reduce topic count to 0.0625x (ablation) | LiteLong · HELMET + RULER, 2B tokens | average score: 49.01 -> 44.80 | Hurt |
| mix FineWeb-Edu with Cosmopedia V2 instead of Cosmopedia V2 only (ablation) | LiteLong · HELMET + RULER | average score: 52.32 -> 61.90 | Improved |
| mix Cosmopedia V2 into FineWeb-Edu (ablation) | LiteLong · HELMET + RULER | average score: 61.29 -> 61.90 | Improved |
| Filter-Reject retention instead of Keep-Accept (ablation) | multi-agent debate · HELMET + RULER | average score: 61.34 -> 61.90 | Improved |
| Filter-Reject retention instead of Keep-Fixed-K-Accept (ablation) | multi-agent debate · HELMET + RULER | average score: 61.54 -> 61.90 | Improved |
| add NExtLong on top of LiteLong | LiteLong · HELMET + RULER average | average score: 61.90 -> 63.04 | Improved |
| add LiteLong to NExtLong | NExtLong · Embedding GPU hours | GPU hours: 928 -> 176 total | Improved |
| train with LiteLong long-context data | LLaMA-3-8B · Short-context tasks average | average score: 62.05 -> 61.92 | No clear effect |
| add LiteLong to NExtLong data before SFT | NExtLong · LongBench v2 | overall: 30.4 -> 30.6 | No clear effect |
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.
- NExtLong — NExtLong: Toward Effective Long-Context Training without Long Documents (ICML 2025) · better on 3 of 4 shared results
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
- Chain of Agents: Large Language Models Collaborating on Long-Context Tasks (NeurIPS 2024 · similarity 0.89)
- Quest: Query-centric Data Synthesis Approach for Long-context Scaling of Large Language Model (ICLR 2025 · similarity 0.88)
- Training with “Paraphrasing the Original Text” Teaches LLM to Better Retrieve in Long-Context Tasks (AAAI 2025 · similarity 0.88)
- Structured Packing in LLM Training Improves Long Context Utilization (AAAI 2025 · similarity 0.88)
- NExtLong: Toward Effective Long-Context Training without Long Documents (ICML 2025 · similarity 0.88)
- BABILong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack (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.
