Synthetic Contrastive Reasoning for Multi-Table Q&A
Quick Answer
This paper shows that A synthetic contrastive reasoning-trace dataset for multi-table Q&A was developed, enhancing models like Qwen3-14B and Mistral-8B with Contrastive Preference Optimization (CPO).
Quick Take
CPO achieved performance gains of 9.7%-16.3% over traditional supervised fine-tuning, with up to 21 percentage points improvement on MMQA, demonstrating the effectiveness of heterogeneous trace generation.
Key Points
- Synthetic dataset enhances multi-table Q&A with reasoning supervision.
- CPO fine-tuning improved model performance by 9.7%-16.3% on average.
- Up to 21 percentage points improvement observed on MMQA benchmarks.
- Heterogeneous trace generators strengthen contrastive signals effectively.
- Evaluations confirm generated pairs are coherent and meaningful.
Paper Resources
Source Excerpt
arXiv:2606. 05382v1 Announce Type: new Abstract: Multi-table question answering requires models to retrieve relevant evidence, link schemas, and perform compositional reasoning across relational tables. Existing multi-table Q&A resources typically provide questions and final answers but lack reasoning supervision that explains how answers are derived.
To address this gap, we construct a synthetic contrastive reasoning-trace dataset for MMQA by generating validated positive traces and plausible negative traces with heterogeneous . We then use the resulting preference pairs to fine-tune open-weight LLMs with Contrastive Preference Optimization (CPO). …
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →RAIL Guard: Closing the Evaluation-to-Remediation Gap in Responsible AI for Agents
RAIL Guard introduces a closed-loop AI pipeline for large language models (LLMs) that evaluates outputs across eight dimensions and iteratively remediates failures, achieving 96.9% convergence compared to 49.1% for traditional block-and-retry methods. The system reduces unsafe agent executions by 33% without impacting task completion and is available as open-source SDKs.