Research map / Large language models / Language, safety and interpretability
TAPEX: Table Pre-training via Learning a Neural SQL Executor
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SQL-execution pre-training | BART-Large · WikiTableQuestions test | denotation accuracy: 38.0 -> 57.5 | Helped |
| add SQL-execution pre-training | BART-Large · WikiSQL-Weak test | denotation accuracy: 85.8 -> 89.5 | Helped |
| add SQL-execution pre-training | BART-Large · SQA test | denotation accuracy: 58.6 -> 74.5 | Helped |
| add SQL-execution pre-training | BART-Large · TabFact test | accuracy: 80.8 -> 84.2 | Helped |
| add multi-task intermediate fine-tuning (ablation) | TAPEX · WikiTableQuestions dev | denotation accuracy: 57.0 -> 57.2 | No clear effect |
| add multi-task intermediate fine-tuning (ablation) | BART · WikiTableQuestions dev | denotation accuracy: 37.2 -> 47.4 | Helped |
| scale pre-training corpus from 0.1M to 5M (ablation) | TAPEX · WikiTableQuestions dev | denotation accuracy: 48.6 -> 57 | Helped |
| scale pre-training corpus from 1.0M to 5M (ablation) | TAPEX · WikiSQL-Weak dev | denotation accuracy: 89.2 -> 89.1 | No clear effect |
| add Medium-difficulty SQL templates to Easy-only corpus (ablation) | TAPEX · WikiTableQuestions dev | denotation accuracy: 41.5 -> 52.1 | Helped |
| add Extra-Hard SQL templates (ablation) | TAPEX · TabFact dev | accuracy: 83.8 -> 83.6 | Hurt |
| pre-train on NL translations instead of SQL queries (ablation) | TAPEX · WikiTableQuestions dev | denotation accuracy: 54.2 -> 52.8 | Hurt |
| pre-train on NL translations instead of SQL queries (ablation) | TAPEX · TabFact dev | accuracy: 83.6 -> 83.7 | No clear effect |
Closest papers by meaning
- Table as a Modality for Large Language Models (NeurIPS 2025 · similarity 0.86)
- TabFlex: Scaling Tabular Learning to Millions with Linear Attention (ICML 2025 · similarity 0.85)
- CARTE: Pretraining and Transfer for Tabular Learning (ICML 2024 · similarity 0.85)
- SQL-R1: Training Natural Language to SQL Reasoning Model By Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- OpenTab: Advancing Large Language Models as Open-domain Table Reasoners (ICLR 2024 · similarity 0.85)
- Programming Every Example: Lifting Pre-training Data Quality Like Experts at Scale (ICML 2025 · similarity 0.85)
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.