SFT on K12-Train improves out-of-domain subjects (Chinese, Humanities Math), suggesting acquisition of a transferable structurally-grounded answer style¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Although K12-Train is synthesized only from mathematics, physics, chemistry, and biology subgraphs, SFT on it attains the highest Chinese score (120.18, +3.33 over DataFlow) and the best Humanities Math score (132.00, +4.08 over DataFlow). → Hao Liang 2026

Evidence¶

Hao Liang 2026¶

Hao Liang, Qihan Lin, Mingrui Chen, Hengyi Feng, Zhaoyang Han, Xiaochen Ma, Zhen Hao Wong, Meiyi Qiang, Linzhuang Sun, Wentao Zhang. (2026). K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs. arXiv preprint. https://arxiv.org/abs/2605.09635 doi:10.48550/arxiv.2605.09635

q2 · i? · design · r2

Cross-subject analysis of GaokaoBench per-subject scores (Table 4) for models trained on K12-Train-Text. The article argues that since these subjects receive no in-domain supervision, the improvement reflects acquisition of a transferable, structurally-grounded answer style rather than content memorization.

"SFT on K12-Train also attains the highest Chinese score (120.18,+3.33over the strongest baseline DataFlow) and the best Humanities Math score (132.00,+4 .08over DataFlow)"

Discussion¶

Related Claims¶

• Under a matched 2,300-sample SFT budget, K12-Train-Text outperforms eight mainstream instruction-tuning corpora on GaokaoBench and EduEval across two base models — related
• K12-Train-Full achieves the best overall performance on three multimodal educational benchmarks, and textual and visual supervision are complementary — related
• LLM teaching effectiveness varies substantially across subject domains, strongest where knowledge points apply directly and weakest in physics and chemistry — related
• Education-oriented instruction tuning consistently improves curriculum grounding, K-12 problem solving, and pedagogical tutoring at every evaluated model scale — a broader claim this one bears on