Kev: Tiny Jev-like family of decision models built on top of Qwen3.5

AI Digest

Kev是基于Qwen3.5的微型决策模型家族，支持多种问题类型，可本地部署运行，适用于客户支持场景，提供概率输出和Web测试环境。

Kev offers tiny decision models built on Qwen3.5, supporting multiple question types with local deployment, probability outputs, and a web playground for customer support scenarios.

Key points

• 提供0.8B/4B/9B参数量模型，支持yes/no/多选/评分问题 Offers 0.8B/4B/9B parameter models with yes/no/multiple-choice/rating support
• 兼容CUDA和Apple Silicon，适配Mac 32GB内存运行 Runs on CUDA and Apple Silicon with Mac 32GB memory compatibility
• 通过TypeSafe SDK对接，支持自定义训练和部署 Integrates with TypeSafe SDK for custom training/deployment
• 包含评估数据和训练代码，可直接测试客户工单 Includes evaluation data and training code for direct testing
• 输出概率分布而非单一标签，提升决策可靠性 Provides probability distributions instead of single labels

Takeaway: Kev通过概率输出和灵活部署，提升客户支持场景的决策准确性。 / Kev enhances customer support decision-making through probabilistic outputs and flexible deployment.

Why it matters 项目提供完整训练框架和测试环境，适合需要自定义AI决策系统的开发者。