本地 LLM 错误排查
粘贴你看到的错误信息。大多数本地 AI 错误都归结为一件事——模型比你的硬件所能承载的更大——每篇指南都会告诉你确切的修复方法。
从你的操作系统开始
全部指南（A–Z） (37)
- "App is damaged and can't be opened" — Gatekeeper on macOS“macOS won't let me open LM Studio, Jan or a llama.cpp binary” · 2026年9月1日macOS
- "Failed to initialize NVML: Driver/library version mismatch"“nvidia-smi errors but I never rebooted” · 2026年9月1日Linux
- "No CUDA-capable device is detected" — how to fix it“My GPU is right there and CUDA says there is no device” · 2026年6月22日WindowsLinux
- "ollama is not recognized as an internal or external command"“The app is installed but the terminal cannot find it” · 2026年9月1日Windows
- "Permission denied" on /dev/kfd — the AMD GPU group fix“Works with sudo, fails as my user” · 2026年9月1日Linux
- AMD GPU not detected / ROCm unsupported (Ollama & llama.cpp)“My Radeon is installed but nothing will use it” · 2026年6月22日LinuxWindows
- Apple Silicon: model won't use the GPU / Metal not engaged“My M-series Mac is running the model on the CPU” · 2026年6月22日macOS
- Can't reach Ollama from another device — Windows Firewall“localhost:11434 works on the PC, nothing else can connect” · 2026年9月1日Windows
- CUDA out of memory — why it happens and how to fix it“It crashes the moment the model loads and says CUDA out of memory” · 2026年6月22日WindowsLinux
- CUDA works on Windows but not inside WSL2“nvidia-smi works in PowerShell, fails in Ubuntu” · 2026年9月1日WindowsLinux
- iPhone model downloads that fail, fill your storage, or stop“The download resets every time I leave the app” · 2026年9月1日iPhone & iPad
- LM Studio: "Failed to load model" (insufficient memory)“LM Studio shows a red "failed to load model" and stops” · 2026年6月22日WindowsLinuxmacOS
- Local LLM outputs gibberish? Wrong chat template — not VRAM“The output is gibberish or it never stops talking” · 2026年7月10日WindowsLinuxmacOS
- Local LLM repeating itself or stuck in a loop — how to fix it“It says the same sentence over and over until I stop it” · 2026年7月10日WindowsLinuxmacOSiPhone & iPad
- MLX won't install, or runs on the CPU“pip can't find mlx, or mlx is slower than llama.cpp” · 2026年9月1日macOS
- Model loads but runs painfully slow — it is on your CPU“It works, but it types slower than I can read” · 2026年6月22日WindowsLinuxmacOS
- Move your models off C: — Ollama and LM Studio on Windows“My C: drive is full of models” · 2026年9月1日Windows
- nvidia-smi stopped working after a kernel update“It worked yesterday, now the GPU is gone” · 2026年9月1日Linux
- Ollama "connection refused" — API not reachable, ports & fixes“Something else on my network can't reach Ollama” · 2026年7月10日WindowsLinuxmacOS
- Ollama on a headless Linux server — disk fills, API unreachable“/ is full and the API only answers on localhost” · 2026年9月1日Linux
- Ollama runs on the CPU on Windows despite an NVIDIA GPU“ollama ps says 100% CPU and my RTX sits idle” · 2026年9月1日Windows
- Ollama: "model requires more system memory than is available"“Ollama refuses to run the model and says I need more memory” · 2026年6月22日WindowsLinuxmacOS
- Out of disk space / failed GGUF download for large models“The download died partway through and my disk is full” · 2026年6月22日WindowsLinuxmacOS
- Out of memory at long context (the KV cache, not the weights)“It loads fine, then dies once the conversation gets long” · 2026年6月22日WindowsLinuxmacOS
- Raising the Metal VRAM ceiling on Apple Silicon“recommendedMaxWorkingSetSize is far lower than my RAM” · 2026年9月1日macOS
- ROCm doesn't support your AMD card — HSA_OVERRIDE and Vulkan“rocminfo sees nothing, or Ollama says no compatible GPUs” · 2026年9月1日Linux
- Running local LLMs on an Intel Mac — what actually works“LM Studio won't install on my 2019 MacBook Pro” · 2026年9月1日macOS
- Secure Boot is blocking your GPU driver module“The driver installed fine and the module still will not load” · 2026年9月1日Linux
- SmartScreen and antivirus are blocking your local AI install“"Windows protected your PC" — or the download just vanished” · 2026年9月1日Windows
- The model crashes your iPhone app, or drops to a smaller one“It quits to the home screen the moment the model loads” · 2026年9月1日iPhone & iPad
- Which GGUF quant should I download? (Q4 vs Q5 vs Q8)“Twenty files on the download page and I don't know which one” · 2026年6月22日WindowsLinuxmacOS
- Windows shared GPU memory makes your model crawl, not crash“It loads fine and then generates at 2 tok/s” · 2026年9月1日Windows
- Your Docker container can't see the GPU“nvidia-smi works on the host, fails inside the container” · 2026年9月1日Linux
- Your laptop runs the model on the integrated GPU, not the RTX“nvidia-smi shows the GPU idle while the model runs” · 2026年9月1日Windows
- Your Mac is swapping to disk instead of running the model“It loaded, then the whole Mac froze” · 2026年9月1日macOS
- Your MacBook is fast for two minutes, then half as fast“tok/s drops by half after a long prompt” · 2026年9月1日macOS
- Your Ollama environment variables are ignored on Linux“I exported OLLAMA_HOST and nothing changed” · 2026年9月1日Linux