# Long-Document QA with Chain-of-Structured-Thought and Fine-Tuned SLMs

arXiv: https://arxiv.org/abs/2603.29232 · HTML: https://mlsanity.com/paper/2603.29232/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 17 of its 53 references that are in the map.
- First posted on arXiv 31 Mar 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 17 of 17)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [From Local to Global: A Graph RAG Approach to Query-Focused Summarization](https://mlsanity.com/paper/2404.16130/) (2024)
- [FinQA: A Dataset of Numerical Reasoning over Financial Data](https://mlsanity.com/paper/2109.00122/) (2021)
- [A Dataset of Information-Seeking Questions and Answers Anchored in Research Papers](https://mlsanity.com/paper/2105.03011/) (2021)
- [InstructUIE: Multi-task Instruction Tuning for Unified Information Extraction](https://mlsanity.com/paper/2304.08085/) (2023)
- [LegalBench-RAG: A Benchmark for Retrieval-Augmented Generation in the Legal Domain](https://mlsanity.com/paper/2408.10343/) (2024)
- [Fin-R1: A Large Language Model for Financial Reasoning through Reinforcement Learning](https://mlsanity.com/paper/2503.16252/) (2025)
- [LLM-Assisted Content Analysis: Using Large Language Models to Support Deductive Coding](https://mlsanity.com/paper/2306.14924/) (2023)
- [PromptNER: Prompting For Named Entity Recognition](https://mlsanity.com/paper/2305.15444/) (2023)
- [YAYI-UIE: A Chat-Enhanced Instruction Tuning Framework for Universal Information Extraction](https://mlsanity.com/paper/2312.15548/) (2023)
- [VisJudge-Bench: Aesthetics and Quality Assessment of Visualizations](https://mlsanity.com/paper/2510.22373/) (2025)
- [DataPuzzle: Breaking Free from the Hallucinated Promise of LLMs in Data Analysis](https://mlsanity.com/paper/2504.10036/) (2025)

## Built on by (most cited first, 3 of 3)

- [DataSpace: Benchmarking Data Agents for Verifiable Analytics over Heterogeneous Workspaces](https://mlsanity.com/paper/2608.03451/) (2026)
- [OCC-RAG: Optimal Cognitive Core for Faithful Question Answering](https://mlsanity.com/paper/2606.00683/) (2026)
- [Beyond Tables: Doc2DB-Bench for Relationally Faithful Document-to-Database Construction](https://mlsanity.com/paper/2608.08459/) (2026)

## Abstract (arXiv)

Large language models (LLMs) are widely applied to data analytics over documents, yet direct reasoning over long, noisy documents remains brittle and error-prone. Hence, we study document question answering (QA) that consolidates dispersed evidence into a structured output (e.g., a table, graph, or chunks) to support reliable, verifiable QA. We propose a two-pillar framework, LiteCoST, to achieve both high accuracy and low latency with small language models (SLMs). Pillar 1: Chain-of-Structured-Thought (CoST). We introduce a CoST template, a schema-aware instruction that guides a strong LLM to produce both a step-wise CoST trace and the corresponding structured output. The process induces a minimal structure, normalizes entities/units, aligns records, serializes the output, and verifies/refines it, yielding auditable supervision. Pillar 2: SLM fine-tuning. The compact models are trained on LLM-generated CoST data in two stages: Supervised Fine-Tuning for structural alignment, followed by Group Relative Policy Optimization (GRPO) incorporating triple rewards for answer/format quality and process consistency. By distilling structure-first behavior into SLMs, this approach achieves LLM-comparable quality on multi-domain long-document QA using 3B/7B SLMs, while delivering 2-4x lower latency than GPT-4o and DeepSeek-R1 (671B). The code is available at https://github.com/HKUSTDial/LiteCoST.