Research map / Large language models / Language, safety and interpretability
Probing How Scalable Table Data Enhances General Long-Context Reasoning
Huaibing Xie, Guoliang Zhao, Yang Liu, Shihan Dou, Siming Huang, Yanling Xiao, Shaolei Wang, Yiting Liu, Cheng Zhang, Shaofan Liu, Pluto Zhou
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 |
|---|---|---|---|
| RL on synthesized table data (TableLong) | DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 40.12 -> 48.36 | Improved |
| RL on TableLong data | DS-R1-Distill-14B · 7 long-context benchmarks average | average accuracy: 36.14 -> 45.07 | Improved |
| RL on TableLong data | DS-R1-Distill-32B · Needle in a Haystack | retrieval accuracy (%): 87.95 -> 99.40 | Improved |
| RL on TableLong data | DS-R1-Distill-32B · LiveCodeBench, AIME 2025, GPQA-Diamond, MultiChallenge (OOD) | accuracy: LCB 46.71 -> 58.68; AIME 60.00 -> 70.00 | Improved |
| RL on TableLong data | Qwen3-32B · MRCR | accuracy: 42.45 -> 42.66 | No clear effect |
| RL on TableLong data | DS-R1-Distill-14B · Loong / GSM-Infinite / MRCR | accuracy: Loong 25.07 -> 35.44; GSM-INF 9.00 -> 9.60 | Mixed |
| use semantics-free (structure only) tables (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 40.12 -> 41.79 (vs 48.36 with full data) | Mixed |
| add cell semantics to table data (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 41.79 -> 48.36 | Improved |
| add visible delimiters to tables (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 47.59 -> 48.36 | No clear effect |
| replace delimiters with noise text (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 48.36 -> 47.38 | No clear effect |
| increase number of involved table cells (0~30 to 300+) (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 46.30 -> 48.36 | Improved |
| increase number of tables per instance (1 to 1~30) (ablation) | TableLong, DS-R1-Distill-32B · 7 long-context benchmarks average | average accuracy: 46.66 -> 48.36 | Improved |
Closest papers by meaning
- NeedleInATable: Exploring Long-Context Capability of Large Language Models towards Long-Structured Tables (NeurIPS 2025 · similarity 0.91)
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding (ICLR 2024 · similarity 0.90)
- InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models (ICLR 2026 · similarity 0.90)
- Longer Context, Deeper Thinking: Uncovering the Role of Long-Context Ability in Reasoning (NeurIPS 2025 · similarity 0.89)
- GSM-$\infty$: How Do your LLMs Behave over Infinitely Increasing Reasoning Complexity and Context Length? (ICML 2025 · similarity 0.89)
- OpenTab: Advancing Large Language Models as Open-domain Table Reasoners (ICLR 2024 · similarity 0.89)
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.
