This work makes the first attempt to build a medalist-level LLM agent for geometry and presents InternGeometry, which solves 44 of 50 IMO geometry problems (2000-2024), exceeding the average gold medalist score, using only 13K training examples, just 0.004% of the data used by AlphaGeometry 2.
This paper conducts a systematic study of unsupervised visual pretraining paradigms that directly leverage visual documents without text extraction, showing that Visual Pretraining is a scalable learner for foundation model intelligence.
Mixed SFT, a single supervised fine-tuning stage that jointly trains on no-CoT and long-CoT data, achieves a clearly higher post-RLVR performance ceiling than next-chunk reasoning RL while requiring over 60 times less training compute.
Yin-Hao Tang, You-Qing Fang, Ya-Nan Sun et al.· 1 citation
SciExplore is introduced, a benchmark designed to evaluate scientific information-seeking and reasoning capabilities of LLMs and agents, revealing substantial performance gaps with performance degrading sharply as task complexity increases and extremely low accuracy on the most challenging structured synthesis tasks.
Yinhao Tang, Youqing Fang, Yanan Sun et al.· Annual Meeting of the Associ...· 1 citation
Existing scaling strategies for Multimodal Large Language Models (MLLMs) typically expand either model parameters or sequential inference computation, incurring substantial memory or latency overhead. More importantly, most existing methods fail to alter the rigid, fixed computation allocation between the Vision Transf...
Yang Yang, Qinyu Zhao, Mou-Xiang Chen et al.· 0 citations
Large language models (LLMs) have achieved remarkable performance on high-school and olympiad-style mathematics, yet their capabilities on advanced mathematics remain poorly understood. Existing benchmarks, however, fall short in both scope and evaluation granularity: they provide limited disciplinary coverage and ofte...
