Unverified50% confidenceFactExact time
Wolf Defender v2结合短样本（256 token）与完整长窗口（2048 token）进行混合训练
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/31/2026
First Seen
Valid until: 11/29/2026
Sources
Related Entities
Related Claims
UnverifiedWolf Defender主模型的真实场景良性特异度从v1的66.85%提升至v2的96.63%62% similarUnverifiedWolf Defender v2的攻击检测能力基本保持原有水平，Qualifire F1为95.14%，Jayavibhav F1为97.84%62% similarUnverified在Gemma 4 12B模型的Token生成测试中，Spark达26.4 tokens/s、Halo 24.6 tokens/s、M4 Pro最快33.8 tokens/s58% similarUnverified根据Chinchilla缩放定律，一个30B模型的最优训练数据量约为600B-1T tokens55% similarUnverifiedMiniMax M3 每个 token 推理时约激活 23B 参数54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/827594API
curl https://kongchang.com/api/v1/knowledge/claims/827594MCP
get_claim(id=827594)