DIY AI: Finetuning Open-Source LLMs Like a Pro!

Abstract:

Open-source small language models (SLMs) like DeepSeek, Phi-3, and Mistral are game-changers in AI accessibility. But how do you get them to work specifically for your use case? Whether you're building a chatbot, a coding assistant, or a niche domain expert, finetuning these models can make a world of difference.

In this session, we’ll demystify the process of finetuning SLMs on consumer hardware and show how to optimize them for real-world applications.

We’ll discuss:

• Choosing the Right Model – Why smaller, open-source LLMs might be better than GPT-4 for your task.
• Data Curation & Preprocessing – Cleaning and structuring your dataset for better results.
• Fine-tuning Techniques – LoRA, QLoRA, and full fine-tuning explained in simple terms.
• Training on a Budget – Running your own fine-tuning experiments without a GPU cluster.
• Evaluating & Deploying – Making sure your model is accurate, efficient, and doesn’t hallucinate nonsense.

This talk is designed for AI enthusiasts, developers, and researchers who want to take control of open-source AI models without burning a hole in their cloud credits. Expect a fast-paced, engaging session with a live demo, practical insights, and a few AI-generated surprises!

Why This Talk?

• The open-source AI community is booming, and small models are the future.
• Many developers use LLMs, but few know how to fine-tune them properly.
• Bangalore’s tech community can benefit from an accessible, practical approach to AI customization.

Takeaways:

By the end of this talk, attendees will know:

✅ How to pick the right open-source model for their task

✅ The simplest ways to fine-tune a model with minimal compute

✅ How to evaluate and deploy their custom-tuned models effectively

Target Audience: College students and working professionals interested in AI, open-source, and machine learning.