Fine-Tune an Open Source LLM with Claude Code/Codex (Hugging Face Model Trainer Skill)

Description

In this video I show how to fine-tune a small open source model with Codex using the Hugging Face model trainer skill and HF Jobs. We train `Qwen/Qwen3-0.6B` into a customer-support ticket router on the Bitext dataset, then compare before/after behavior with JSON/schema routing metrics. I also cover setup prerequisites, run monitoring, and how to deploy the result as a lightweight service. Connect with me 🤗 - X: https://x.com/_alejandroao - LinkedIn: https://www.linkedin.com/in/alejandro-ao/ Links - Code: https://github.com/huggingface/hub-tutorials/blob/main/notebooks/06-hf-model-trainer-skill.ipynb - Model Trainer Skill: https://github.com/huggingface/skills/tree/main/skills/hugging-face-model-trainer - Release Blog Post: https://huggingface.co/blog/hf-skills-training 🤓 Topics Covered - Codex + Hugging Face Jobs workflow - Qwen3 ticket-router fine-tuning setup - Before/after eval and cost breakdown ⏱️ Timestamps 0:00 Intro and goal 1:05 Ticket-routing app framing 2:14 Model and dataset selection 5:04 Notebook setup and pre-finetune baseline 9:40 Prompt and training requirements 13:50 Submit training job 14:45 Monitor training outputs 17:18 Cost review 18:10 Results and before/after demo