Start fine-tuning your own multilingual gpt-oss-20B model today.
Earlier this month, we made available OpenAI’s open‑source model gpt‑oss on Azure AI Foundry and Windows AI Foundry. Today, you can fine-tune gpt‑oss‑20b using Managed Compute on Azure — available in preview and accessible via notebook.
As open-weight models gain traction, developers are increasingly looking for ways to customize and deploy them efficiently. Large language models like gpt-oss-20B are revolutionizing AI applications. But what if you could fine-tune them for your domain or language with just a single H100?
Why Fine-tune gpt-oss-20B?
gpt-oss-20B is a powerful, open-source model. Fine-tuning enables:
- Domain-specific adaptation (e.g., customer support, education)
- Multilingual reasoning (English, Spanish, French, Italian, German)
- Efficient training with LoRA adapters—only 1% of parameters are updated
With just a few clicks, you can fine-tune gpt-oss-20B for multi-lingual chain-of-thought reasoning. The platform handles distributed training, resource scaling, and output packaging.
Image – Select gpt-oss-20b
Fine-tuning Use Cases & Applications
From multilingual chatbots and assistants that deliver seamless global interactions, to domain-specific reasoning tools tailored for legal, financial, and technical decision-making, organizations can unlock new efficiencies and insights.
|
Category |
Use Case |
|
Multilingual Chatbots & Assistants |
Real-time support, translation, and conversational agents across languages |
|
Domain-Specific Reasoning Tools |
Legal, medical, financial, and technical copilots for reasoning and decision support |
|
Global Customer Support |
Automated ticket triage, sentiment analysis, and escalation routing |
|
Developer & Data Tools |
Code generation, tool-use agents, and data science copilots |
Why use Managed Compute?
Managed Compute is a deployment option within Azure AI Foundry Models that lets you run large language models (LLMs), SLMs, HuggingFace models and custom models fully hosted on Azure infrastructure. Azure Managed Compute is a powerful deployment option for models not available via standard (pay-go) endpoints. It gives you:
- Custom model support: Deploy open-source or third-party models
- GPU flexibility: tested on Standard_NC96ads_A100_v4 and Standard_ND96isr_H100_v5
- Detailed control: Configure inference servers, protocols, and advanced settings
- Full integration: Works with Azure ML SDK, CLI, Prompt Flow, and REST APIs
- Enterprise-ready: Supports VNet, private endpoints, quotas, and scaling policies
This setup is ideal for developers who want to fine-tune and deploy models without managing infrastructure.
Getting Started
Currently, fine-tuning gpt-oss-20B is available only via notebook with Azure ML. There’s no UI support in the Azure yet (UI enablement coming soon), so developers work directly with code in this scenario. This notebook walks through:
- Setting up Azure ML compute
- Configuring LoRA adapters
- Training with multilingual datasets
- Deploying for chat completion tasks
➡️ Get started with this GitHub Notebook
Learn More with these Resources
🧠 Get started with Azure AI Fine-tuning on Microsoft Learn
👩💻 Learn more about gpt-oss on Azure AI Foundry and Windows AI Foundry
👋 Continue the conversation on Discord
Microsoft