Mistral AI发布了Shieldstral 1.0 3B，一个专为策略适应性设计的开源权重安全分类器。与依赖固定危害类别的传统模型不同，Shieldstral使用自然语言问题进行内容审核，允许操作员在推理时定义自定义策略。这种方法使该模型在文本和多模态安全基准测试中都取得了强劲的性能，其表现可与更大的模型相媲美，同时能在配备16GB显存的单GPU上高效运行。 AI
影响 为各种应用提供灵活的本地内容审核，可能减少对第三方供应商的依赖。
排序理由 前沿实验室模型发布，附带系统卡。
在 Mastodon — mastodon.social 阅读 →
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
<p>Mistral AI has released Shieldstral 1.0 3B, an open-weights, policy-adaptive multimodal safety classifier that frames content moderation as a single yes/no question instead of a fixed harm taxonomy. Operators supply the policy as a plain-language query at inference time and ge…
Mistral AI has released Shieldstral 1.0 3B, an open-weights safety classifier that adapts to different policies at runtime. Rather than using a fixed harm taxonomy, it treats content moderation as a yes/no question. Achieves 84.9% F1 on text safety benchmarks while running on a s…
🧠 # Mistral # AI ha rilasciato # Shieldstral 1.0 3B, un modello open-weight progettato per la moderazione e la sicurezza dei contenuti. 👉 I dettagli: https://www. linkedin.com/posts/alessiopoma ro_mistral-ai-shieldstral-share-7492449329505337344-xahm/ ___ ✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶…
Mistral's Shieldstral is a 3B open model that performs safety checks via natural language questions, matching larger models in benchmarks. Operators can define their own criteria and run it locally. Source: The Decoder AI https:// the-decoder.com/mistrals-open- model-shieldstral-…
Mistral veröffentlicht Shieldstral, ein 3B Open-Weight Safety-Modell, das feste Kategorien durch natürliche Sprache ersetzt. Es schlägt rechenintensivere Klassifikatoren bei gleicher Benchmark-Leistung – relevant für lokale Inferenz. https:// the-decoder.de/mistrals-shield stral-…