Guides

It will not run

Work down the list. This is not a catalog of every machine.

  1. The runner is not installed. Go back to Install and use one official download.
  2. The tag or the filename is wrong. Use the name on the official library page, then the matching model page on this site.
  3. The file does not fit in memory. Open the model page and read the computed Q4_K_M + 8,192 KV total. If that number is larger than the RAM or VRAM you have, pick a smaller model or a shorter context. Do not start with a 70B.
  4. The GPU is not seen. The run may have fallen back to the CPU, or it may have stopped. That is not a speed result.
  5. A prompt format is required. gpt-oss needs the harmony format from its model card. Qwen3 can spend the first reply in a thinking trace unless you turn that off.
  6. A license or a gate is blocking the official download. The model page links the license. A gated repo stays gated.

When accounts exist, a useful help note includes the runner and version, the exact model tag, the machine’s RAM and GPU, and the exact error. Never paste keys.

Official docs, not a random thread: Ollama quickstart, LM Studio downloads, llama.cpp.