Skip to content

ollama_win.md

Latest commit

 

History

History
82 lines (60 loc) · 4.24 KB

ollama_win.md

File metadata and controls

82 lines (60 loc) · 4.24 KB
Raw

Static Badge

Installing Ollama on Windows

  1. Grab the installer from the Ollama download page.
  2. Double-click the ollamasetup.exe and follow the installation prompts (you've done this before).
  3. Verify the installation by opening your terminal (CMD Prompt, Putty...) and type ollama --version.
ollama version is 0.9.6

Run a Model

  1. Open/reopen your preferred terminal application
  2. Type ollama run gemma3:1b, since it's the first time you've run this, it will download the LLM model.
  3. Let's test it out with a picture analysis. My test is a picture featuring dueling force-wielding cats. Here's the start of the output from the gemma:1b model, which returned immediately: ollama run gemma3:1b
Okay, let's break down what we can gather about the image you provided:

**Overall Impression:**

The image is a highly stylized, almost cartoonish, depiction of a cat and a dog engaged in a challenging and playful "duel." It's a visual metaphor for a struggle of opposing forces or a competition.  The style leans 
heavily into a slightly exaggerated, slightly dreamlike aesthetic.

Close Ollama by typing /bye.

One of the cats is apparently a dog. Per Google "The Gemma3 1B size models are text only and do not support image input. The model was unable to fully understand the picture and what we experienced was a form of AI hallucination.

To increase accuracy let's step it up to the gemma:4b model. Note: This will take longer as it's a larger LLM and we're running on non-AI optimized hardware.

  1. Type ollama run gemma3:4b, and then after the download completes, re-enter the same prompt as before.
Okay, let's break down this wonderfully surreal image! 

**What it depicts:**

This is a digitally manipulated image depicting two orange tabby cats engaged in a lightsaber duel against a backdrop of a bright, cloud-filled sky with a prominent rainbow.  Each cat is wielding a red lightsaber, 
and beams of red light are shooting between them.

This took about 30 seconds to process, but far more accurate to my picture!

Bonus challenge

When running ollama on your laptop, it runs an http service on localhost port 11434, so you can send queries to ollama via another service. Let's try that!

curl http://localhost:11434/api/generate -d \
'{
  "model": "gemma3:4b",
  "prompt": "what can you tell me about this picture? /Users/Keanu/AI-stepbystep/lab1/dueling-cats-2.png"
}'

Huzzah! That works. But...the output is not, uh, great:

{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.058712Z","response":"Okay","done":false}
{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.08365Z","response":",","done":false}
{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.110131Z","response":" let","done":false}
{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.135686Z","response":"'","done":false}
{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.162662Z","response":"s","done":false}
{"model":"gemma3:4b","created_at":"2025-04-14T22:13:44.188051Z","response":" analyze","done":false}

Ollama streams the response word-by-word by default. To fix this, we can set stream to false:

curl http://localhost:11434/api/generate -d \
'{
  "model": "gemma3:4b",
  "prompt": "what can you tell me about this picture? /Users/Keanu/AI-stepbystep/lab1/dueling-cats-2.png",
  "stream": false
}'

That looks better, not great, but better. Remember we're calling the API so it's not meant for normal consumption but at least we can distil the output easier.

Conclusion

Ollama is the foundation for many AI tools you'll use in later labs. Now you have your first model is running - boom, you're officially in the AI game! - Play around and get comfortable with the basics, then explore and see what's next!