How to import a model into Ollama

Importing a model brings external GGUF or Safetensors weights into Ollama so the model can be run through the normal CLI and API. The Modelfile describes whether Ollama should load a model file, a directory of weights, or an adapter.

The import source and base model must match the weights. A LoRA or Safetensors adapter built for one base model can behave unpredictably if the Modelfile points at a different base.

Use a small test import before moving large production weights. Model conversion and quantization can consume significant disk and memory.

1. Place the GGUF or Safetensors source in a local directory.

   $ ls /srv/models/custom-model
   model.gguf

2. Create a Modelfile that points at a GGUF model file.

   FROM /srv/models/custom-model/model.gguf

3. Create the Ollama model name from the Modelfile.

   $ ollama create example/custom-model -f Modelfile
   transferring model data
   writing manifest
   success

4. Run the imported model.

   $ ollama run example/custom-model "Return OK."
   OK

5. Inspect the model before publishing or using it in automation.

   $ ollama show example/custom-model
     Model
       architecture        llama
   ##### snipped #####