| Tool | What it is | Source |
|---|---|---|
| llama.cpp | Open-source C/C++ inference engine; defines the GGUF format; CPU, CUDA, Metal, Vulkan and other backends | llama.cpp: quantize README (bits/weight table) |
| Ollama | Command-line/server tool for downloading and running models | Ollama |
| LM Studio | Desktop app with a GUI and local server | LM Studio |
Hardware impact: the same GGUF model needs similar memory in each; differences come from default context length, KV-cache settings and backend. Performance comparison on identical hardware: TBD.