Fine-tuning Gemma 3, Qwen3, Llama 4, Phi 4 and Mistral Small - Advanced Guide

Learn advanced fine-tuning techniques for the latest language models including Gemma 3, Qwen3, Llama 4, Phi 4, and Mistral Small in this comprehensive tutorial from Trelis Research. Discover when fine-tuning is appropriate, how to prepare effective datasets, and the differences between using Unsloth and Transformers libraries. Follow along with practical demonstrations on running fast evaluations with vLLM in Jupyter notebooks, implementing fine-tuning with Unsloth, and using the Transformers library for model customization. The tutorial covers essential decision points like model selection, general fine-tuning best practices, and provides complete setup instructions for hands-on implementation. Access the repository and additional resources through Trelis.com/ADVANCED-fine-tuning to apply these techniques to your own projects.

Syllabus

0:00 Fine-tune Gemma Llama Qwen Phi Mistral
0:27 Video Overview
1:35 When to do fine-tuning?
3:22 How to prepare a fine-tuning dataset?
6:24 Choosing unsloth versus transformers
11:07 How to run fast evaluations with vLLM
12:33 What model to fine-tune?
14:27 General fine-tuning tips
16:04 Fine-tuning notebooks and setup: trelis.com/advanced-fine-tuning
19:11 Running fast evaluations with vLLM in a jupyter notebook
37:56 Fine-tuning with Unsloth
1:09:02 Fine-tuning with transformers
1:20:24 Conclusion