CTO and head of engineering at x-Hub.io, with over 16 years of IT experience. In 2010, I co-founded the first developer community in Morocco: MoroccoJUG. Two years later, I co-founded the JMaghreb conference, which became the largest developer conference in the Middle East and Africa. JMaghreb was rebranded in 2015 when we joined the Devoxx franchise, solidifying its position as the most prestigious developer conference in the MEA region.
I also played a role in creating the Java Wave community in Quebec City, Canada, where I lived for nine years. During my time there, I spoke at multiple local events.
My interest in the AI world began in 2017 when we built a semantic search engine for Canada's largest research funding organization. This engine utilized word2vec to create embeddings, long before the emergence of vector databases. I'm also the creator of "laarbi," an AI assistant that understands the Moroccan dialect.
Throughout my career, I've contributed to various open-source projects including the JSF Mojarra implementation, JSR 346, and Reactivemongo. I was also honored as the Outstanding Adopt-a-JSR Participant at the 11th JCP Annual Awards.
In this 30-minute quickie, I'll creating a personalized AI chatbot using the LLaMA 3 language model. We'll explore how to fine-tune this powerful model to mimic my own conversation style, using my personal WhatsApp chat history spanning 9 years.
Key points we'll cover:
- Setting up the environment using Google Colab and Unsloth for efficient fine-tuning
- Preparing and preprocessing 9 years of WhatsApp conversation data
- launch fine-tuning process.
- Test the trained model trained on 50 steps
- Compare to a model trained for 1 epochs on the same data.
- Demonstrated overfitting with a model trained for 5+ epochs.
Throughout the session, I'll share my screen to guide you through the Google Colab notebook, providing real-time insights into the code and processes involved.
Searching for speaker images...