Data-Juicer Sandbox: A Feedback-Driven Suite for Multimodal Data-Model Co-development

The emergence of large-scale multimodal generative models has drastically advanced artificial intelligence, introducing unprecedented levels of performance and functionality. However, optimizing these models remains challenging due to historically isolated paths of model-centric and data-centric developments, leading to suboptimal outcomes and inefficient resource utilization. In response, we present a novel sandbox suite tailored for integrated data-model co-development. This sandbox provides a comprehensive experimental platform, enabling rapid iteration and insight-driven refinement of both data and models. Our proposed "Probe-Analyze-Refine" workflow, validated through applications on state-of-theart LLaVA-like and DiT-based models for image-to-text and text-to-video tasks, yields significant performance boosts, such as topping the VBench leaderboard. We also uncover fruitful insights gleaned from exhaustive benchmarks, shedding light on the critical interplay between data quality, diversity, and model behavior. All codes, datasets and models are open-sourced to foster future progress. Under review as a conference paper at ICLR 2025 ModelTrainExecutorFactory LLaVA-like TrainExecutor … Sora-like TrainExecutor ModelEvaluatorFactory MM-Bench Evaluator V-Bench Evaluator … Analyze Stats Correlation