OmniEdu capability-oriented instruction-tuning corpus¶
Product or Programme · All products and programmes
Evidence · 3 claims (3 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 3 claims rest on one study
Description¶
The OmniEdu corpus is a curated, audited, diverse instruction-tuning dataset for education-focused language models, created by Hao Liang et al. from educational datasets and resources.
Components¶
- OmniEdu capability-oriented instruction-tuning corpus and six-stage data pipeline: A reproducible data-construction pipeline that curates an education-specific pool of approximately 1.34M examples drawn from more than 100 datasets and educational resources into 60,951 high-quality examples containing about 12.0M supervised response tokens. The six stages combine deterministic cleaning and evaluation decontamination, LLM-assisted semantic auditing and rewriting with 0-100 usability scoring, task-specific quality scoring, k-center greedy diversity selection over BGE-M3 embeddings under token budgets, and pedagogical instruction assignment. Together with 9,048 general-purpose examples, the final corpus contains 69,999 examples and 15.96M supervised response tokens, with provenance and audit metadata retained throughout. (Hao Liang et al. (2026))
Claims¶
- Education-oriented instruction tuning consistently improves curriculum grounding, K-12 problem solving, and pedagogical tutoring at every evaluated model scale [+M]
- Education-oriented tuning improves authentic K-12 problem solving, with OmniEdu-27B's MDK12-Bench open-question correctness rising from 62.08% to 78.77% [+W]
- Specializing models for K-12 education does not require a broad loss of general capability, with MMMU-Pro accuracy increasing at all three scales [+W]
Related Products and Programmes¶
Key Sources¶
- Hao Liang, Qihan Lin, Meiyi Qiang, Linzhuang Sun, Hengyi Feng, Mingrui Chen, Sizhe Qiu, Wentao Zhang. (2026). OmniEdu: Open Foundation Models for Learning and Teaching. https://arxiv.org/abs/2609.23088
Page metadata
| Field | Value |
|---|---|
type |
product |
id |
omniedu-capability-oriented-instruction-tuning-corpus |
title |
OmniEdu capability-oriented instruction-tuning corpus |
description |
The OmniEdu corpus is a curated, audited, diverse instruction-tuning dataset for education-focused language models, created by Hao Liang et al. |
status |
review |
product_kind |
dataset |
generated |
by process:settle-candidates on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.