Vision-Based Fine-tuning Gemma 3 LLM with Unsloth on Google Colab

NOTE: When defining the instruction at 5:13, it's better to have a period (.) at the end.
So instead of "Convert this image to JSON", it's better to write "Convert this image to JSON." (notice the full-stop at the end of this string). It is fixed in the code in the link below.

Code: https://colab.research.google.com/drive/1VMnGLcamszwwQ6MXujWdda5f0YOa5X9_?usp=sharing

Gemma 3 Modelfile: https://github.com/ollama/ollama/issues/9783#issuecomment-2726499020

Unsloth Docs: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Gemma3_(4B)-Vision.ipynb

Previous fine-tuning video: https://youtu.be/iAlAWRVAiCg
Hey guys, in this video, you're gonna learn how to fine-tune the Gemma 3 Vision model using the Unsloth framework. We will be using Google Colab for this tutorial.
Timestamps:
0:00 Intro
0:42 Building our dataset
3:40 Saving dataset to disk
6:34 Formatting our dataset
7:08 Installing Unsloth
7:29 Importing Gemma 4B
9:58 Preparing for fine-tuning
12:52 Fine-tuning the model
13:21 Performing inference
17:11 Saving LoRA adapters
17:50 Testing LoRA adapters
19:28 Saving in Float16
20:20 Saving as GGUF file
21:00 Installing and starting Ollama
21:48 Creating Modelfile
23:32 Testing the model
25:00 Outro

Thanks for watching! I am open to any suggestions and reviews :)

Видео Vision-Based Fine-tuning Gemma 3 LLM with Unsloth on Google Colab канала Hussain Arif