大模型
arxiv.org ·
QVAC Genesis III：面向高效语言模型预训练的大规模高质量开放合成 STEM 语料库
该 arXiv 预印本提出 QVAC Genesis III，一个 191.43B token、覆盖 19 个 STEM 领域、多难度与多教学风格的合成语料库，面向 token 预算受限的边缘 AI 与端侧小模型预训练。其构建采用双重生成策略：以弱边缘规模学生模型为信号做定向教师蒸馏，将学生失败转化为纠错解释，将成功扩展为覆盖所有选项的对比式推理；同时提出 LLM-as-a-parser 评测协议，从自由文本输出中抽取最终答案并统计准确率与答案有效率。作者用 1.7B 参数模型从零训练做对照消融，称在 ARC、GPQA Diamond 和 MMLU STEM 上优于使用开源合成语料 Cosmopedia-v2 训练的模型及公开的 Cosmo-1B，ARC-E 最高提升 28.57%、ARC-C 最高提升 21.35%，有效答案率最高 99.45%。
AI推荐理由 arXiv 预印本，摘要给出明确规模（191.43B token、19 个 STEM 领域）、方法（以弱边缘学生模型的失败/成功做教师蒸馏与对比式选项推理）和可验证的对照实验（1.7B 参数从零训练，对比 Cosmopedia-v2 与 Cosmo-1B，ARC-E +28.57%、ARC-C +21.35%，有效答案率最高 99.45%）。新信息在于面向边缘/端侧小模型的高效 STEM 合成语料与 LLM-as-a-parser 评测协议。局限：仅为预印本，未提供同行评审、数据开放许可细节与第三方复现，且对比基线有限，收益幅度需独立验证。
78
正文
Computer Science > Artificial Intelligence
arXiv:2609.19513 (cs)
[Submitted on 17 Sep 2026 (v1), last revised 23 Sep 2026 (this version, v2)]
Title:QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training
Authors:Davide Vitabile, N. Ranjan, Akshay Nambiar, Kamal K. Gupta, Amril Nazir
View a PDF of the paper titled QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training, by Davide Vitabile and 4 other authors
View PDF
HTML (experimental)
Abstract:High-quality pre-training data is a critical bottleneck for educational and STEM-specific language models targeting edge AI and on-device deployment where token budgets are tightly constrained. While major organizations train ever-larger models on private corpora, the open ecosystem lacks STEM-focused synthetic datasets that deliver high per-token learning value efficiently for small models. To address this gap, we introduce QVAC Genesis III, a 191.43B-token, STEM-focused multi-domain synthetic corpus covering 19 domains across several difficulty levels and different educational styles. QVAC Genesis III is built via a dual generation strategy that performs targeted teacher distillation using a weak edge-scale student model as signal: the student's failures are converted into corrective explanations, while its successes are expanded into contrastive option-level reasoning over all answer choices. We further introduce an LLM-as-a-parser evaluation protocol that extracts final answers from free-form outputs and tracks both accuracy and answer validity. To validate the effectiveness of our QVAC Genesis III data, we conduct controlled from-scratch ablations with 1.7B-parameter models, showing that models trained with QVAC Genesis III consistently outperform both models trained with the open-source synthetic corpus Cosmopedia-v2 and the publicly released Cosmo-1B model across ARC, GPQA Diamond, and MMLU STEM benchmarks, achieving up to +28.57% on ARC-E and +21.35% on ARC-C, while reaching a Valid Answer Rate of up to 99.45%.
Subjects:
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as:
arXiv:2609.19513 [cs.AI]
(or
arXiv:2609.19513v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.19513
Focus to learn more
arXiv-issued DOI via DataCite
Submission history
From: Amril Nurman Nazir [view email]
[v1]
Thu, 17 Sep 2026 00:04:09 UTC (1,511 KB)
[v2]
Wed, 23 Sep 2026 06:16:38 UTC (772 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled QVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training, by Davide Vitabile and 4 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.AI
new
|
recent
| 2026-09
Change to browse by:
cs
cs.LG
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)