Import Your Own Model into Ollama: Safetensors Folders, Single and Split GGUF Files

By Web Asset House Studio · October 29, 2026

Note: Some future links on this site may be affiliate links. We never claim hands-on testing; guides are researched from the public sources listed in each article.

How we wrote this: A plain-English summary of Ollama's official Importing a Model documentation, checked October 29, 2026. Check each model's license before you use or share it (see open model licenses).

If a model isn't in the Ollama library, perhaps a fine-tune you downloaded from Hugging Face, you can import it yourself. Every import uses the same pattern: a Modelfile with a FROM line, then ollama create.

From Safetensors weights

Create a file named Modelfile that points at the folder holding the weights:
```
FROM /path/to/safetensors/directory
```
If the Modelfile is inside that folder, FROM . works.
From the Modelfile's folder, build it:
```
ollama create my-model
```
Test it:
```
ollama run my-model
```

From a single GGUF file

FROM /path/to/file.gguf

Then run ollama create my-model.

From a split GGUF (several shards)

Big GGUF models often come as several files. The docs say to keep the original shard file names and use a wildcard that matches all of them:

FROM /path/to/model-*.gguf

Quantize before importing

The docs are explicit: Ollama does not quantize GGUF models during import. If you want a smaller version, quantize it first with a GGUF tool such as llama.cpp's llama-quantize. See llama.cpp for beginners and what is a GGUF file.

Add your own settings

The same Modelfile can set a system prompt and parameters (for example PARAMETER num_ctx 8192). See our Modelfile guide.

Sharing (optional)

The docs also explain pushing a model to ollama.com, which needs a free account and your Ollama public key: rename it with ollama cp mymodel youruser/mymodel, then run ollama push youruser/mymodel. Imported models stay private on your machine unless you push them.