Train AI models with Unsloth Studio without coding
Exclusive UPDF discount link:
https://updf.com/youtube/globalcobots
Step-by-step tutorial for fine-tuning Gemma 4 with Unsloth Studio using QLoRA, without programming during training.
In this video, I train Gemma 4 locally with a 16GB RTX 5080, using my own Spanish dataset on the European Artificial Intelligence Regulation. We'll see how to configure LoRA, learning rate, epochs, batch size, warmup, evaluation loss, checkpoints, and how to interpret the training curves to avoid overfitting.
Video resources:
Unsloth Studio tool installation.
https://unsloth.ai/
Gemma 4 Model:
https://ollama.com/library/gemma4
Dataset created:
https://huggingface.co/datasets/hugor...
Script to create the dataset from PDF:
https://huggingface.co/hugoramallo/ge...
Trained repository/model (this was a code training):
https://huggingface.co/hugoramallo/ge...
Video Chapters:
00:00 Introduction to No-Code Fine-Tuning
01:00 Organizing Documents with UPDF
02:35 Configuring Gemma 4 and QLoRA
03:47 First Training: Configuration Hyperparameter Models
05:49 Result 1: Overfitting Analysis
06:50 Second Training: Adjusting Dropout and Learning Rate
07:44 Result 2: Stable Curve and Improved Generalization
08:32 Real-World Test: Evaluating Model Responses
09:50 Why Specialist Models Are the Future