DESIGNER: Design-Logic-Guided Multidisciplinary Data Synthesis for LLM Reasoning

TL;DR

This work tackles the scarcity of high-quality, multidisciplinary reasoning data for large language models by introducing DESIGNER, a design-logic-guided data synthesis pipeline. It leverages book and web corpora to extract a reusable library of design logics, which are used in a two-stage retrieval-and-generation process to synthesize millions of challenging, cross-disciplinary questions (DLR-Book and DLR-Web) across 75 disciplines. The authors validate the data by extensive SFT experiments on Qwen3 and Llama3 models, showing substantial improvements in multidisciplinary reasoning that in some cases surpass official post-training baselines. The study also provides a thorough analysis of difficulty, diversity, data quality, and ablations, underscoring the utility of design logics for controllable, scalable, and transferable reasoning data creation. The work is poised to enhance cross-domain reasoning capabilities in LLMs and offers a reproducible design-logic library and processing pipeline for future research.

Abstract

Large language models (LLMs) have achieved remarkable success in many natural language tasks but still struggle with complex, multi-step reasoning, particularly across diverse disciplines. Existing reasoning datasets often lack disciplinary breadth, reasoning depth, and diversity, as well as guiding principles for question synthesis. We propose DESIGNER: a DESIGN-logic-guidEd Reasoning data synthesis pipeline that leverages naturally available, extensive raw documents (e.g., book corpus and web corpus) to generate multidisciplinary challenging questions. We introduce the concept of "design logic" and instruct LLMs to mimic human educators' question-creation process, enabling the automated synthesis of large-scale, high-difficulty questions. We use LLMs to reverse-engineer and abstract over 120,000 design logics from existing questions across various disciplines. By matching these design logics with source documents, we are able to generate reasoning questions with controllable question types and difficulty levels. Using this pipeline, we synthesized two large-scale reasoning datasets that span 75 disciplines: DLR-Book (3.04 million questions from the book corpus) and DLR-Web (1.66 million questions from the web corpus). Data analysis indicates that the questions synthesized by our method exhibit greater difficulty and diversity compared to those in the baseline datasets. We validate our synthesized data through supervised fine-tuning (SFT) on the Qwen3 and Llama3 model families. Our data substantially enhances their multidisciplinary reasoning capabilities, outperforming existing datasets. Notably, by applying SFT on the base versions of these models using only our data, we even surpass their official final models that have undergone the full post-training process.

Figures (25)

• Figure 1: Left: The procedure by which human experts construct questions reflects a "Design Logic": a systematic sequence of deliberate steps that transforms fundamental knowledge points into complex, context-rich questions requiring multi-stage reasoning. Right: DESIGNER emulates this process by matching logics to raw corpus and synthesizing diverse, multidisciplinary questions.
• Figure 2: The Design-Logic-Guided Multidisciplinary Data Synthesis Pipeline.
• Figure 3: Difficulty distributions of questions across different datasets and benchmarks.
• Figure 4: Discipline distribution across different datasets.
• Figure 5: Performance scaling with synthesized data size. Benchmark accuracy increases steadily with data scale.