Uploaded October 2023 | Updated September 2026, 2 weeks ago

---

Fine-tuning large models doesn't have to be complicated and expensive. In this tutorial, I provide a step-by-step demonstration of the fine-tuning process for a LLaMA 2 7B model. Utilizing a pre-existing script sourced from the TRL library, the configuration is set to leverage the QLoRA algorithm from the Hugging Face PEFT library. The training procedure is executed on a modest AWS GPU instance (g5.xlarge), optimizing cost-effectiveness through the utilization of EC2 Spot Instances, resulting in a total cost of just a few dollars.

⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️

- Blog: huggingface.co/blog/4bit-transformers-bitsandbytes
- Model + code + training log: huggingface.co/juliensimon/llama2-7b-qlora-openassistant-guanaco
- Dataset: huggingface.co/datasets/timdettmers/openassistant-guanaco
- Amazon EC2 G5 instances: aws.amazon.com/ec2/instance-types/g5

Thumbnail generated with Stable Diffusion XL : "Clint Eastwood western style riding a lama"

Follow me on Medium at julsimon.medium.com or Substack at https://julsimon.substack.com.