Learn how to fine-tune a small language model with Unsloth Desktop, evaluate it against the base model, export it as GGUF, and call it from a Python application. This is the beginner walkthrough I wanted when I started learning fine-tuning. We use a small Qwen model and a finance-operations task: turn an incident into a JSON response with a category, severity, and next action. I walk through the whole process, including a configuration mistake, the correction, and the actual evaluation results. A completed training run does not automatically mean you have a better model. The comparison is what tells you where to improve next. You'll learn how to: • Define a small task you can measure • Structure examples and separate training from validation data • Establish a baseline before training • Understand LoRA, QLoRA, and the settings used in this experiment • Spot the difference between three optimizer steps and three epochs • Compare the base and fine-tuned models using the same evaluation cases • Export to GGUF and run the model locally • Call the model through an OpenAI-compatible API from Python CHAPTERS 00:00 What we're building 01:47 The full fine-tuning journey 02:34 What fine-tuning actually changes 03:42 Define the task and output format 04:43 Structure the training examples 05:20 Establish a baseline and automate evaluation 06:41 Training, validation, and test data 08:06 LoRA and QLoRA explained 09:42 Choose the model and datasets in Unsloth 12:26 Configure the training run 16:30 Gradient accumulation explained 17:16 Start training 18:03 The mistake: steps versus epochs 20:36 Read the training charts 22:40 Compare responses in chat 24:18 Run the automated evaluation 25:54 Compare the scores and plan the next iteration 28:09 Merge and export to GGUF 31:10 Load and test the exported model locally 33:05 Call the model from Python 36:36 The model still needs a system around it 37:32 How to approach your first fine-tune JOIN MY COMMUNITY TO GET ACCESS TO THE RESOURCES The linked v0.1 model and dataset are the earlier published experiment introduced in the video. The tutorial includes a separate training run and evaluation. This is a learning experiment, not a production finance system. Subscribe for more practical AI engineering tutorials as I keep working through fine-tuning, evaluation, and the systems around models. #FineTuning #Unsloth #AIEngineering