Fine-tuning LLMs with PEFT and LoRA - Gemma model & HuggingFace dataset

In this video, I give a code walkthrough of finetuning the latest Gemma 2B parameter model from Google in the HuggingFace ecosystem.

I have created a new Databricks-dolly-mini dataset derived from the Databricks-dolly-15k dataset. The newly created dataset is then used to fine-tune Gemma 2b to visualize the training progress and the results.


⌚️ ⌚️ ⌚️ TIMESTAMPS ⌚️ ⌚️ ⌚️
0:00 - Intro
1:24 - Preliminaries & Installation
2:44 - Run inference on pre-trained Gemma-2b
4:44 - Motivation for Parameter Efficient Fine-tuning (PEFT)
7:58 - Create a custom dataset for finetuning
11:58 - Tips for fine-tuning (parameters)
19:29 - Supervised Fine-tuning
20:30 - Training visualization & Interpretations
23:04 - Conclusion


RELATED LINKS
Colab Notebook: https://github.com/ai-bites/generativ...
Google's Gemma: https://blog.google/technology/develo...
Databricks Dolly 15k dataset: https://huggingface.co/datasets/datab...
New Databricks Dolly mini dataset: https://huggingface.co/datasets/ai-bi...
LoRA paper: https://arxiv.org/abs/2106.09685
My LoRA video:    • LoRA (Low-rank Adaption of AI Large L...  
My QLoRA video:    • QLoRA paper explained (Efficient Fine...  


MY KEY LINKS
YouTube:    / @aibites  
Twitter:   / ai_bites​  
Patreon:   / ai_bites​  
Github: https://github.com/ai-bites​


WHO AM I?
I am a Machine Learning researcher/practitioner who has seen the grind of academia and start-ups equally. I started my career as a software engineer 16 years ago. Because of my love for Mathematics (coupled with a glimmer of luck), I graduated with a Master's in Computer Vision and Robotics in 2016 when the now happening AI revolution started. Life has changed for the better ever since.

#machinelearning #deeplearning #aibites