What is the difference between full fine-tuning, LoRA, and p-tuning on Databricks?

- - Certifications
- - Learning Paths
- - Databricks Product Tours
- - Get Started Guides

- - Get Started Resources
- - Events
- - Support FAQs
- - Technical Blog
- - Community Articles
- - Announcements
- - DatabricksTV
- - Product Platform Updates

- - Private Groups
  - Princeton Life Sciences Databricks User Group
- - Skills@Scale

- - Databricks Community Champions
- - Khoros Community Forums Support (Not for Databricks Product Questions)
- - Databricks Community Code of Conduct

Generative AI

Explore discussions on generative artificial intelligence techniques and applications within the Databricks Community. Share ideas, challenges, and breakthroughs in this cutting-edge field.

How are these three methods different?

Full Fine-Tuning
LoRA
P-Tuning

1 REPLY 1

Hi @Suheb

Summary Table:

Method What’s Tuned Speed/Cost Flexibility Use Case

Full Fine-Tuning	All model weights	Slow/High	Maximum	Custom tasks, large data
LoRA	Small adapter layers	Fast/Low	High	Efficient adaptation
p-tuning	Prompt embeddings	Fastest/Low	Limited	Prompt-based tasks