Llama Guard local deployment tutorial: run Llama Guard 3, ShieldGemma and Qwen3Guard with Ollama, with VRAM math, moderation code and a model table.
Does Qwen need a jailbreak locally? Most refusals are over-refusal, template or quantization issues. A diagnosis flow, lawful fixes and risks, with no prompts.
Red-teaming a local LLM: compare garak, PyRIT, promptfoo, Inspect and HarmBench, with authorization, metrics, Wilson interval math and a fix loop.
Llama Guard 本地部署教程：用 Ollama 运行 Llama Guard 3、ShieldGemma 与 Qwen3Guard，含显存估算、审核代码结构与选型表。
Qwen 本地部署需要破甲吗？多数人遇到的其实是过度拒答、模板错误或量化问题。本文给出判断流程、合规的解决办法、评估方法和风险，不提供越狱提示词。
本地模型红队测试入门：对比 garak、PyRIT、promptfoo、Inspect 与 HarmBench，讲清授权隔离、指标、置信区间算例与修复闭环。
Refusal direction ablation explained: the single-direction finding, weight orthogonalization, what KL divergence means, side effects and comparisons.
拒绝方向消融原理详解：Arditi 论文的单方向发现、权重正交化怎么做、KL 散度的含义、副作用，以及与微调、DPO、提示词越狱的对比。
Windows WSL2 Docker local LLM setup: install WSL 2, share the NVIDIA driver, run Ollama with –gpus, set memory limits, and know when native is simpler.
AnythingLLM tutorial: install the desktop app or Docker, connect Ollama, pick an embedder, build workspaces, tune Query mode and fix ‘no relevant information’.
