If you haven’t tried it already, Ollama is a great tool built on top of llama.cpp that makes it easier to run small language models (SLMs) like Phi-3 and Llama3-8B on your own machine, even if your personal computer has no GPU or has an ARM chip. Ollama provides both a command-line interface to chat with the language model, as well as an OpenAI-compatible chat completion endpoint.

What if your personal computer can’t run Ollama for some reason, like if you’re using a ChromeBook or iPad without the ability to install? GitHub Codespaces to the rescue! Codespaces is a way to open any GitHub repository in the browser, inside a web-based VS Code running a containerized development environment, all customizable via a devcontainer.json file.

We can add Ollama to the Codespace for a repository by adding this community-created feature in devcontainer.json:

“features”: {
“ghcr.io/prulloac/devcontainer-features/ollama:1”: {}
},

Once we open a repository with that feature added, we can open the terminal of the Codespace in the browser and run a model from the Ollama models catalog:

We can also call that Ollama server programmatically, either via its standard endpoint or via its OpenAI-compatible endpoint using an OpenAI SDK:

import openai

client = openai.OpenAI(
base_url=”http://localhost:11434/v1″,
api_key=”nokeyneeded”,
)
response = client.chat.completions.create(
model=”phi3:mini”,
messages=[
{“role”: “system”, “content”: “You are a helpful assistant.”},
{“role”: “user”, “content”: “Write a haiku about a hungry cat”},
],
)
print(response.choices[0].message.content)

Ollama Python Playground

To make it easy for you to get started using Ollama with Python, open the Codespace for this repository:

https://github.com/pamelafox/ollama-python-playground/

That repo includes the Ollama feature, OpenAI SDK, a notebook with demonstrations of few-shot and RAG, and a Python script for an interactive chat. It’s designed to be a helpful resource for teachers and students who want a quick and easy way to get started with small language models.

Ollama C# Playground

If you want to use Ollama from .NET instead, open this C# playground in Codespaces:

https://github.com/elbruno/Ollama-CSharp-Playground

That repo also includes sample code for common tasks with the SLMs. Bruno put together a video walking through it as well:

Phi-3 CookBook

We’ve also added the Ollama feature to the Phi-3Cookbook repository:

https://github.com/microsoft/Phi-3CookBook

If you open that repo in a GitHub Codespace, then you can use Ollama while reading through the guides. Of course, there is a limit to what you can do in a Codespace, due to the lack of a GPU and general resource constraints. Ollama does a great job optimizing for those scenarios, but that cookbook also contains Jupyter notebooks that use the transformers package, which is not optimized for the non-GPU case. When I tried to run the Phi-3 inference notebook, it took my Codespace a full 1.5 hours to complete the inference. :face_screaming_in_fear:

We hope that this lowers the barrier even more for everyone interested in trying out small language models! Let us know in the comments if you’ve added Ollama support to any repositories or if you have other ways that you like to experiment with small language models.

​Microsoft Tech Community – Latest Blogs –Read More