A controlled study proving that the temporal organization (curriculum) of multimodal data is a first-order variable in balancing reasoning vs. OCR capabilities.
March 31, 2026
Original Paper
Data Organization Matters in Multimodal Instruction Tuning: A Controlled Study of Capability Trade-offs
arXiv · 2603.27744
The Takeaway
Reveals that simple data mixing is suboptimal; specific curriculum training strategies provide significantly better trade-offs between general understanding and structured reasoning. This offers a direct recipe for optimizing multimodal instruction tuning pipelines.
From the abstract
Recent multimodal large language models (MLLMs) perform strongly on general visual understanding, diagram and chart reasoning, and document-centric perception. However, these abilities are learned from heterogeneous supervision sources with very different task structures and learning demands, and the effect of their temporal organization during training remains underexplored. We study whether data organization affects the trade-off among general understanding, structured reasoning, and fine-grai