The lesson was built using core resources from OpenAI and Microsoft Foundry as references for the terminology and tutorials. Here is a non-comprehensive list for your own self-guided learning journeys. Every link below points to current, supported material.
1. Primary Resources
| Title/Link | Description |
|---|---|
| Fine-tuning with OpenAI Models | Fine-tuning improves on few-shot learning by training on many more examples than can fit in the prompt - saving costs, improving response quality, and enabling lower-latency requests. Get an overview of fine-tuning from OpenAI. |
| When to use Microsoft Foundry fine-tuning | Understand what fine-tuning is (concept), why you should consider it, what data to use, and how to measure quality - plus when SFT, DPO, or RFT is the right fit. |
| Customize a model with fine-tuning | The end-to-end how-to (process) for fine-tuning in Microsoft Foundry using the portal, the OpenAI / Foundry Python SDK, or the REST API - covering data prep, training, checkpoints, and deployment. |
| Continuous fine-tuning | The iterative process of selecting an already fine-tuned model as the base model and fine-tuning it further on new sets of training examples. |
| Fine-tuning with tool (function) calling | Fine-tuning your model with tool-calling examples improves output - more accurate, consistent, similarly-formatted responses using fewer prompt tokens. |
| Fine-tuning models: Microsoft Foundry guidance | Look up which models can be fine-tuned, the methods they support (SFT / DPO / RFT), and the regions where they're available. |
| Fine-tuning overview: techniques and modalities | Compare the three training techniques (SFT, DPO, RFT) and the two modalities (serverless vs. managed compute), with guidance on choosing a base model and getting started. |
| Tutorial: Fine-tune a model in Microsoft Foundry | Create a sample dataset, prepare for fine-tuning, run a fine-tuning job on a currently supported model such as gpt-4.1-mini, and deploy the fine-tuned model on Azure. |
| Tutorial: Fine-tune models with serverless API deployments | Tailor open and partner models (Phi, Llama, Mistral, and more) to your datasets using a low-code, UI-based workflow in Microsoft Foundry. |
| Tutorial: Fine-tune Hugging Face models on Azure Databricks | Fine-tune a Hugging Face model with the transformers library on a single GPU using Azure Databricks and the Hugging Face Trainer. |
| Training: Fine-tune a foundation model with Azure Machine Learning | The Azure Machine Learning model catalog offers many open-source models you can fine-tune. Part of the Azure ML Generative AI Learning Path. |
| Tutorial: Azure OpenAI fine-tuning with Weights & Biases | Track and analyze fine-tuning runs on Azure with W&B. Extends the OpenAI fine-tuning guide with Azure-specific steps and experiment tracking. |
2. Secondary Resources
This section captures additional resources worth exploring that we didn't have time to cover in the lesson. Use them to build your own expertise around this topic.
| Title/Link | Description |
|---|---|
| OpenAI Cookbook: Data preparation and analysis for chat model fine-tuning | Preprocess and analyze a chat dataset before fine-tuning: check for format errors, get basic statistics, and estimate token counts (and cost). Pairs with the OpenAI fine-tuning guide. |
| OpenAI Cookbook: Fine-tuning for Retrieval Augmented Generation (RAG) with Qdrant | A comprehensive example of fine-tuning OpenAI models for RAG, integrating Qdrant and few-shot learning to boost performance and reduce fabrications. |
| OpenAI Cookbook: Fine-tuning GPT with Weights & Biases | Use W&B to track model training and fine-tuning. Read their OpenAI Fine-Tuning guide first, then try the Cookbook exercise. |
| Hugging Face Tutorial: How to Fine-Tune LLMs with Hugging Face TRL | Fine-tune open LLMs using Hugging Face TRL, Transformers, and datasets: define a use case, set up a dev environment, prepare a dataset, fine-tune, evaluate, and deploy. |
| Hugging Face: AutoTrain Advanced | A no-code / low-code library from Hugging Face for fine-tuning many model types. Run it in your own cloud, on Hugging Face Spaces, or locally via GUI, CLI, or YAML config. |
| Unsloth: Fine-tuning LLMs Guide | An open-source framework that streamlines local LLM fine-tuning and reinforcement learning, with ready-to-use notebooks. |