How to Request GPU Quota Increase for Mistral-7B LoRA Training After Upgrading to Pay-As-You-Go?

Question

I recently upgraded my Azure subscription from Free Trial to Pay-As-You-Go, expecting to be able to request GPU quota increases. However, when I navigate to the quota request page, I only see vCPU quotas available—there is no option to request vGPUs.

I specifically need access to A100 or other high-memory GPUs for fine-tuning Mistral-7B with LoRA. Could someone clarify:

1. How do I get GPU options to appear in my quota request page?
2. Is there an additional step needed after upgrading to Pay-As-You-Go to enable vGPU requests?
3. If quota increase isn't an option, what’s the best way to access GPUs on Azure for training large models?

I’ve also considered Azure Machine Learning Studio, but I need dedicated GPU resources. Any guidance would be appreciated!

Answer 1

Hi Dr Kalu Goodness Chibueze,
Greetings & Welcome to the Microsoft Q&A forum! Thank you for sharing your query.

It sounds like you're encountering some common issues when transitioning to a Pay-As-You-Go subscription and trying to request GPU quotas. Here are some steps and tips to help you navigate this process:
How to Get GPU Options to Appear in Your Quota Request Page

Check Subscription and Region:

• Ensure your subscription is correctly set to Pay-As-You-Go.
• Verify that the region you are operating in supports the GPU types you need (e.g., A100).

Request GPU Quota Increase
The GPU resource may be requested also via Azure Portal -> Your subscription -> Usage + quotas blade. Select the target VM size and click on "Request Adjustment" to proceed with request. The GPU resource will depend on your subscription type as well as resource availability in the region.

Additional Steps After Upgrading to Pay-As-You-Go

Sometimes, it takes a while for the subscription upgrade to reflect fully. Ensure your subscription status is active and not pending.
Accessing GPUs on Azure for Training Large Models

• Azure Machine Learning Studio provides managed compute clusters with GPU support. You can create a compute cluster with the desired GPU type and use it for training.
• Follow the distributed GPU training guide for detailed steps on setting up and using GPUs for large model training
• For dedicated GPU resources, consider using Azure's Virtual Machine Scale Sets with GPU instances. This allows you to scale up and down based on your training needs.

I hope this information helps.