MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning

• URL: http://arxiv.org/abs/2507.16812v1
• Date: Tue, 22 Jul 2025 17:59:03 GMT
• Title: MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning
• Authors: Run-Ze Fan, Zengzhi Wang, Pengfei Liu,
• Abstract summary: We present TextbookReasoning, an open dataset featuring truthful reference answers extracted from 12k university-level textbooks.<n>We introduce MegaScience, a large-scale mixture of high-quality open-source datasets totaling 1.25 million instances.<n>Our experiments demonstrate that our datasets achieve superior performance and training efficiency with more concise response lengths.
• Score: 24.72798058808192
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Scientific reasoning is critical for developing AI scientists and supporting human researchers in advancing the frontiers of natural science discovery. However, the open-source community has primarily focused on mathematics and coding while neglecting the scientific domain, largely due to the absence of open, large-scale, high-quality, verifiable scientific reasoning datasets. To bridge this gap, we first present TextbookReasoning, an open dataset featuring truthful reference answers extracted from 12k university-level scientific textbooks, comprising 650k reasoning questions spanning 7 scientific disciplines. We further introduce MegaScience, a large-scale mixture of high-quality open-source datasets totaling 1.25 million instances, developed through systematic ablation studies that evaluate various data selection methodologies to identify the optimal subset for each publicly available scientific dataset. Meanwhile, we build a comprehensive evaluation system covering diverse subjects and question types across 15 benchmarks, incorporating comprehensive answer extraction strategies to ensure accurate evaluation metrics. Our experiments demonstrate that our datasets achieve superior performance and training efficiency with more concise response lengths compared to existing open-source scientific datasets. Furthermore, we train Llama3.1, Qwen2.5, and Qwen3 series base models on MegaScience, which significantly outperform the corresponding official instruct models in average performance. In addition, MegaScience exhibits greater effectiveness for larger and stronger models, suggesting a scaling benefit for scientific tuning. We release our data curation pipeline, evaluation system, datasets, and seven trained models to the community to advance scientific reasoning research.

Related papers

• HiPerRAG: High-Performance Retrieval Augmented Generation for Scientific Insights [72.82973609312178]
  HiPerRAG is a workflow to index and retrieve knowledge from more than 3.6 million scientific articles.<n>At its core are Oreo, a high- throughput model for multimodal document parsing, and ColTrast, a query-aware encoder fine-tuning algorithm.<n>HiPerRAG delivers robust performance on existing scientific question answering benchmarks and two new benchmarks introduced in this work.
  arXiv  Detail & Related papers  (2025-05-07T22:50:23Z)
• SciHorizon: Benchmarking AI-for-Science Readiness from Scientific Data to Large Language Models [36.724471610075696]
  We propose SciHorizon, a comprehensive assessment framework designed to benchmark the readiness of AI4Science.<n>First, we introduce a generalizable framework for assessing AI-ready scientific data, encompassing four key dimensions: Quality, FAIRness, Explainability, and Compliance.<n>We present recommendation lists of AI-ready datasets for Earth, Life, and Materials Sciences, making a novel and original contribution to the field.
  arXiv  Detail & Related papers  (2025-03-12T11:34:41Z)
• SciClaimHunt: A Large Dataset for Evidence-based Scientific Claim Verification [7.421845364041002]
  We introduce two large-scale datasets, SciClaimHunt and SciClaimHunt_Num, derived from scientific research papers.<n>We propose several baseline models tailored for scientific claim verification to assess the effectiveness of these datasets.<n>We evaluate models trained on SciClaimHunt and SciClaimHunt_Num against existing scientific claim verification datasets to gauge their quality and reliability.
  arXiv  Detail & Related papers  (2025-02-14T08:34:26Z)
• MMSci: A Dataset for Graduate-Level Multi-Discipline Multimodal Scientific Understanding [59.41495657570397]
  We present a comprehensive dataset compiled from Nature Communications articles covering 72 scientific fields.<n>We evaluated 19 proprietary and open-source models on two benchmark tasks, figure captioning and multiple-choice, and conducted human expert annotation.<n>Fine-tuning Qwen2-VL-7B with our task-specific data achieved better performance than GPT-4o and even human experts in multiple-choice evaluations.
  arXiv  Detail & Related papers  (2024-07-06T00:40:53Z)
• SciDMT: A Large-Scale Corpus for Detecting Scientific Mentions [52.35520385083425]
  We present SciDMT, an enhanced and expanded corpus for scientific mention detection. The corpus consists of two components: 1) the SciDMT main corpus, which includes 48 thousand scientific articles with over 1.8 million weakly annotated mention annotations in the format of in-text span, and 2) an evaluation set, which comprises 100 scientific articles manually annotated for evaluation purposes.
  arXiv  Detail & Related papers  (2024-06-20T22:03:21Z)
• A Comprehensive Survey of Scientific Large Language Models and Their Applications in Scientific Discovery [68.48094108571432]
  Large language models (LLMs) have revolutionized the way text and other modalities of data are handled. We aim to provide a more holistic view of the research landscape by unveiling cross-field and cross-modal connections between scientific LLMs.
  arXiv  Detail & Related papers  (2024-06-16T08:03:24Z)
• SciRIFF: A Resource to Enhance Language Model Instruction-Following over Scientific Literature [80.49349719239584]
  We present SciRIFF (Scientific Resource for Instruction-Following and Finetuning), a dataset of 137K instruction-following demonstrations for 54 tasks. SciRIFF is the first dataset focused on extracting and synthesizing information from research literature across a wide range of scientific fields.
  arXiv  Detail & Related papers  (2024-06-10T21:22:08Z)
• MASSW: A New Dataset and Benchmark Tasks for AI-Assisted Scientific Workflows [58.56005277371235]
  We introduce MASSW, a comprehensive text dataset on Multi-Aspect Summarization of ScientificAspects. MASSW includes more than 152,000 peer-reviewed publications from 17 leading computer science conferences spanning the past 50 years. We demonstrate the utility of MASSW through multiple novel machine-learning tasks that can be benchmarked using this new dataset.
  arXiv  Detail & Related papers  (2024-06-10T15:19:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.