Ant financial discloses model training data to meet EU AI transparency standards

Ant Financial's inclusionAI has launched a public repository on HuggingFace, releasing summaries of training content for models like Ling, Ring, and Ming. These documents, which do not include model weights or datasets, detail the origins of models from Ling-2.0 to Ling-3.0, Ring-2.0 to Ring-2.6, and Ming-Omni. The summaries adhere to the EU AI Act's Article 53(1)(d), a standardized format for transparency compliance. Each document is specific to a model version, such as Ling-2.0, and does not automatically cover other models in the series. The repository includes versions like Ling-3.0-flash-VL and UniAudio-16B-A3B, with each summary marked by its version and update date. Ant Financial clarifies that these documents do not constitute certification or endorsement of all models. The official notes that the dates in the PDFs refer to document versions, not release dates, and recommends checking commit records for actual changes. This move aligns with global efforts to make training transparency a regulatory requirement, showcasing Ant's compliance strategy and model development insights.