inclusionAI Releases SingGuard Series, First Version SingGuard-0.8b
Original title:inclusionAI/SingGuard-0.8b
AI summary
inclusionAI released the SingGuard series on HuggingFace, with the first version being SingGuard-0.8b. The model takes safety policies as runtime inputs rather than fixed training-time classifications, supporting safety evaluation across six scenarios: text, image, image-text, multilingual, query side and response side. SingGuard uses a dynamic reasoning process, supporting fast first-token routing to output immediate safety signals and continuing to generate more precise judgments when deeper reasoning is needed. Across six benchmark categories covering multimodal safety, image safety, text query safety, text response safety, multilingual query safety and multilingual response safety, SingGuard achieves average SOTA performance. The model natively supports the chat message input format of Transformers and vLLM.
Key Takeaways
- inclusionAI released the SingGuard series on HuggingFace,with the first version being SingGuard-0.
- The model takes safety policies as runtime inputs rather than fixed training-time…
- SingGuard uses a dynamic reasoning process,supporting fast first-token routing to output immediate…
- Across six benchmark categories covering multimodal safety,image safety,text query safety,text…
- The model natively supports the chat message input format of Transformers and vLLM.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 65
蚂蚁这个 SingGuard 把安全策略从固定的分类器变成运行时动态输入,对出海应用的合规审核有实际价值,尤其是多模态场景,我觉得能省掉很多重训成本。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: inclusionAI Releases SingGuard Series, First Version SingGuard-0.8b. June 24, 2026. https://new.nat6.net/en/news/cmref8c5000edihgvnpxf2wfc/