Model Mondays S1:E5 - Hands-on With Fine Tuning Models

Join us for Model Mondays – a weekly show where we round up all the latest AI model news and put one model in the spotlight for a deeper dive.

In this episode, we’re joined by Rashaud Savage as we dive into the word of fine-tuning for AI model customization. We put the spotlight on Mistral models - and discuss how fine-tuning works with Mistral, when to use it - and the workflow to go from data-set to deployment. And we’ll look at core tools and workflows for performant fine-tuning.

Session Resources:
https://aka.ms/model-mondays
https://aka.ms/model-mondays/discord
https://aka.ms/model-mondays/chat

Explore The Models
https://aka.ms/model-mondays/collection

Watch The Replay
https://developer.microsoft.com/en-us/reactor/events/25356/

Transcript

1. Apr 07, 2025 | 1:30pm EST https://aka.ms/model-mondays Hands on with Fine Tuning Models
2. Senior AI Advocate https://aka.ms/model-mondays/RSVP Nitya Narasimhan, PhD Sharmila Chockalingam Senior Product Marketing Mgr. Apr 07, 2025 – 10:30am PT | 1:30pm ET Welcome to Model Mondays
3. Sharmila Chockalingam 5 Key Things To Know This Week! Model Mondays Highlights
4. ❶ · Try Model Leaderboard https://ai.azure.com/explore/models/leaderboard Model Leaderboard Use the leaderboard to see how different model perform for different criteria
5. ❷ · Try Llama 4 Models https://ai.azure.com/explore/models?selectedCollection=meta Meta's Llama4 Scout & Maverick on Azure AI Build more personalized multimodal experiences with 10Million context length.
6. ❸ · Try Agent Framework Read the Blog Post Agent Framework An extension of Azure AI Foundry’s open-source kit Semantic Kernel, specifically designed to simplify the orchestration of multi-agent systems.
7. ❹ · Try VS Code Extension Azure AI Foundry Extension for VS Code Build, test and deploy AI agents and LLMs in VS code. Read the Blog Post
8. ❺ · Try AI Red Teaming Agent AI Red Teaming Agent Agent systematically probes AI models to uncover safety risks, integrating Azure AI Foundry’s robust evaluation systems Read the Blog Post
9. Hands-on with Fine-Tuning Models Rashaud Savage Product Manager (AI Core), Microsoft What is Fine-Tuning and why do we need it? How can we fine-tune Mistral models? What are the pros & cons here?
10. ❶ · What is Fine Tuning? What is Fine-Tuning Fine-tuning is the process of taking a pre-trained model and adjusting it to perform better on a specific task or dataset. This process involves using additional training data that is more closely related to the target task, so the model can learn nuances specific to that task without starting from scratch.
11. ❷ · Why should you use it? Fine-Tuning: Why & When Why When Improve Performance on Specific Tasks When You Have a Specific Task Faster and More Efficient Training When You Have Limited Data Adapt to Specific Data When You Want to Improve Accuracy or Relevance Leverage Large Pre-Trained Models When Updating to Reflect New Data or Trends
12. ❸ · What can Mistral do? Unlocking the Potential of Mistral Models • High-Performance Mistral AI focuses on creating high-performance and efficient models. This allows businesses and individuals to integrate advanced AI technology less expensively and with more customization. • Strong Focus on Responsible AI • Mistral Mistral AI emphasizes building models with responsible AI principles. This focus can be important for companies that need to ensure that their AI solutions align with ethical standards. • Advanced Reasoning Mistral models demonstrates state-of-the-art mathematical and reasoning capabilities within their respective size categories. • Agent-centric Mistral models possesses top-tier agentic capabilities
13. ❹ · How to FT with Mistral How To Fine-Tune With Mistral Models • Sign in to Azure AI Foundry. • Choose the Mistral model you want to fine-tune from Azure AI Foundry model catalog. • On the model's Details page, select fine-tune. • Select the project in which you want to fine-tune your models. • On the fine-tune wizard you have to subscribe your project for the associated model offering (for example, Mistral-Nemo) from Azure Marketplace. o Each project has its own subscription to the model offer, which allows you to control and monitor spending. • Select Subscribe and fine-tune. • Step-by-Step Guides: o Fine-tune models using serverless APIs in Azure AI Foundry portal - Azure AI Foundry | Microsoft Learn o Fine Tune Mistral Models on Azure AI Foundry | Microsoft Community Hub
14. ❺ · Som FT best practices Fine-Tuning Best Practices • Choose a relevant pre-trained model that aligns with your task. • Prepare and preprocess your dataset carefully. • Split your data into training, validation, and test sets effectively. • Tune hyperparameters (especially learning rate and batch size). • Prevent overfitting with techniques like early stopping and regularization. • Monitor model performance regularly and analyze errors. • Save model checkpoints to safeguard your progress. • Evaluate on real-world data to ensure the model is truly effective.
15. Hands-on with Local AI Development Learn how the AI Toolkit for Visual Studio Code enables you to download and run models locally, with cutting edge tools to support testing, fine-tuning & deployment Rong Lu, Principal PM Manager - Microsoft
16. Join the #model-mondays channel https://aka.ms/model-Mondays/chat Q&A Office hours every Friday on Discord – bring questions, share your projects!
17. Watch the Livestream Join the Conversation Explore the Resources Mon 1:30pm ET Microsoft Reactor https://aka.ms/model-mondays/RSVP Fri 1:30pm ET Azure AI Discord https://aka.ms/model-mondays/Discord Watch or Star GitHub Repo https://aka.ms/model-monday #ModelMondays