AI Paper Atlas

Research map / Large language models / Agents, code and math

Beyond Text-to-SQL: Can LLMs Really Debug Enterprise ETL SQL?

Jing Ye, Yiwen Duan, Yonghong Yu, Xinpei Zhao, Yang Gao, Xing Chen

ICML 2026 · Large language models › Agents, code and math · approach: Code generation benchmarks · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
fine-tune Qwen-2.5-Coder-7B on in-domain SQL debugging data (vanilla SFT)Qwen-2.5-Coder-7B · Squirrel-SyntaxGM score: 8.96 -> 30.70Improved
fine-tune the 7B model with in-domain SFT instead of using the 32B modelQwen-2.5-Coder · Squirrel-SyntaxGM score: 20.26 (32B) -> 30.70 (7B + SFT)Improved
fine-tune Qwen-2.5-Coder-7B with vanilla SFTQwen-2.5-Coder-7B · Squirrel-SemanticGM score: 7.75 -> 15.70Improved
mask loss on unchanged tokens (DM-SFT) instead of full-loss SFT (ablation)Qwen-2.5-Coder-7B SFT · Squirrel-SyntaxGM score: 30.70 -> 33.18Improved
predict only diff segments (Diff-SFT) instead of full SQL (ablation)Qwen-2.5-Coder-7B SFT · Squirrel-SyntaxGM score: 30.70 -> 22.81Mixed
use reasoning models rather than non-reasoning modelsLLMs · Squirrel-Syntax and Squirrel-SemanticGM score / MBImproved
use Text-to-SQL specialized model (OmniSQL-32B) for debuggingOmniSQL-32B · Squirrel-SyntaxGM score: 6.40 GM despite 87.6% on SpiderHurt
Kimi-K2 main agent plus Qwen3-Coder sub-agent with TQS feedbackKimi-K2 · Squirrel-SyntaxEM / GM score: EM 14.07 -> 23.24; GM 27.72 -> 28.78Improved
GPT-4o main agent plus Qwen3-Coder sub-agentGPT-4o · Squirrel-SyntaxGM score: GM 4.69 -> 14.37Mixed
use multi-round agent modificationagent systems · Squirrel-SyntaxMB score: Qwen3-Coder-480B 61.62 -> 49.89 (Kimi-K2 + Qwen3-Coder agent)Hurt
give explicit error messages (Syntax vs Semantic task)LLMs · SquirrelGM score: Claude-4-Sonnet 32.17 -> 36.46Improved
train for more steps beyond about 400 on vanilla SFT (ablation)Qwen-2.5-Coder-7B SFT · Squirrel-SyntaxperformanceNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy