DataClaw0: 从原始流中智能定制多模态数据
DataClaw0: Agentic Tailoring Multimodal Data from Raw Streams
- University of Chinese Academy of Sciences（中国科学院大学）
- Shenzhen University of Advanced Technology（深圳理工大学）
- Tsinghua University（清华大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
提出DataClaw0模型，通过两阶段流水线将生成语义合成锚定于确定性事实锚点，结合SFT与GRPO实现复杂数据精炼意图的对齐，首个数据精炼基准验证其高效性。
AI中文摘要：
大规模非结构化多模态流存在高“数据熵”，阻碍了高效的人类知识获取和高质量的AI后训练。现有的被动标注范式严重依赖启发式规则或通用VLM，成本高、单调且无法解锁原始数据中嵌入的深层过程逻辑。我们将数据处理提升为可学习的能力，提出向智能数据定制的范式转变，主动精炼和结构化数据以对齐多样化的用户和下游意图。为了克服训练这种高阶能力的数据稀缺瓶颈，我们设计了一个两阶段流水线，将生成语义合成锚定于确定性事实锚点，生成覆盖五个核心物理和数字领域的大规模数据集。在此基础上，$\text{DataClaw}_0$-9B模型协同监督微调（SFT）与组相对策略优化（GRPO），实现了与复杂精炼和定制意图的稳健对齐。为了系统量化这一能力，我们构建了$\text{DataClaw}_0$-val，这是首个专门用于数据精炼的基准。关键的是，我们采用下游后训练作为最终的验证试金石。在视频生成、真实世界VQA和GUI导航上的评估证实，$\text{DataClaw}_0$提供高信息密度的定制数据，促进模型在有限训练数据下高效适应新任务。项目页面：this https URL
英文摘要：
Raw multimodal streams are abundant but noisy, redundant, and unaligned with any particular training objective. Turning them into supervision today means either brittle heuristics or repeatedly querying a proprietary vision-language model, a cost that recurs with every new sample. We ask whether this conversion can instead be learned once and reused, and formalise intent-conditioned Data Tailoring: given a raw stream and a high-level intent, a model must return schema-aligned, evidence-grounded training instances. Training DataClaw0 at 4B, 9B and 27B, we find that whether five heterogeneous domains should share one model depends on capacity. A jointly trained model is worse than per-domain experts at the two smaller scales and better at the largest, placing the crossover near 18B parameters. Matched-data comparisons, in which the joint model sees exactly the same data per domain as that domain's expert, attribute the reversal to cross-domain transfer rather than to data volume: it gains most where a domain is data-poor, and recovers 59\% of in-domain performance on domains withheld from training entirely. Downstream post-training reproduces this ordering on GUI navigation, action video generation and spatio-temporal VQA, and the joint configuration is also the cheaper to deploy, serving one model instead of five. Github: https://github.com/vancyland/DataClaw0
