Jagle: Building a Large-Scale Japanese Multimodal Post-Training Dataset for Vision-Language Models

TL;DR

Jagle is the largest Japanese multimodal dataset for vision-language models, enabling improved multilingual performance and broad task coverage through diverse data collection and generation strategies.

Contribution

This work introduces Jagle, a large-scale Japanese multimodal dataset created from heterogeneous sources, enhancing Japanese VLM training and performance.

Findings

A 2.2B model trained on Jagle outperforms existing Japanese VLMs.

Combining Jagle with FineVision improves English performance.

Jagle achieves strong results on ten Japanese evaluation tasks.

Abstract

Developing vision-language models (VLMs) that generalize across diverse tasks requires large-scale training datasets with diverse content. In English, such datasets are typically constructed by aggregating and curating numerous existing visual question answering (VQA) resources. However, this strategy does not readily extend to other languages, where VQA datasets remain limited in both scale and domain coverage, posing a major obstacle to building high-quality multilingual and non-English VLMs. In this work, we introduce Jagle, the largest Japanese multimodal post-training dataset to date, comprising approximately 9.2 million instances across diverse tasks. Rather than relying on existing VQA datasets, we collect heterogeneous source data, including images, image-text pairs, and PDF documents, and generate VQA pairs through multiple strategies such as VLM-based QA generation, translation, and text rendering. Experiments demonstrate that a 2.2B model trained with Jagle achieves strong performance on Japanese tasks, surpassing InternVL3.5-2B in average score across ten Japanese evaluation tasks and approaching within five points of Qwen3-VL-2B-Instruct. Furthermore, combining Jagle with FineVision does not degrade English performance; instead, it improves English performance compared to training with FineVision alone. To facilitate reproducibility and future research, we release the dataset, trained models, and code.