The authors built isolated, executable database sandboxes that expose real schemas, values, query results and runtime errors.
Training SQL agents inside their execution harness improves accuracy
HarnessSQL teaches compact models to inspect, query and revise within live database environments rather than learning only final SQL answers.
Big Tech
Haolin Yang · Jipeng Zhang · Jian Xie · Shuaishuai Gong · Sirui Han · Yike Guo
Hong Kong University of Science and Technology · Microsoft Research · Tsinghua University · University of Macau
Research Digest··2 min read
Yang and colleagues address a mismatch between conventional text-to-SQL training and deployment, where agents must interact repeatedly with a database before submitting an answer.
Why this paper
From Microsoft Research and 3 others
In one line
Training SQL agents inside their execution harness via SFT and RL substantially improves execution accuracy on interactive database benchmarks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 8B)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§