CONTROLLED DATA
Leidos Proprietary - US Citizens ONLY
The information contained herein is proprietary to Leidos, Inc. It may not be used, reproduced, disclosed, or exported without the written approval of Leidos.

Describe steps for start/deploy a LLM project on Azure

Step-by-step guide

This is a simple version. If you want a complete how-to go https://learn.microsoft.com/en-us/azure/cognitive-services/openai/overview

  1. log in https://portal.azure.com/#home
  2. Click on Azure Open AI service, which may be on top of the page, if not, search "Azure Open AI"in the search bar
  3. click on the page below

              . 

      4. Fill out the form, provide or create resource group, and instance Name. Other options should has only one default value.

      5. Select network configuration. Recommend to choose "selected networks" and put IP address where you are going to access the service. Keep most other options as it is.

 


  ( Later if you need to add/modify IPs, go to the corresponding resource page, on the left panel, find 'Networking')

 7.  Once a resource is created, you can deploy model within it.  For example, for resource 'augment', this is what it looks like. Click on manage deployments will take you to the Azure AI studio,

where you can deploy models, try it out in play ground or use the model via API calls.

8. For each resource, you can get its key and endpoint from the UI above (left panel under Keys and Endpoint), information on endpoint format can be found here:

 https://learn.microsoft.com/en-us/azure/cognitive-services/openai/reference

7. Afterwards, following the tutorials below for select, customize, and deploy models. e.g., 




Stay on top of the cost and here is how to check cost:

From Azure Home, go to Azure OpenAI, then select the project (you should see 2 currently, RFP-assistant and Cyborg)Once you click into a project, on the left sidebar you will see “Cost analysis”From there you can define the range in which you wish to see the costs.




The information contained herein is proprietary to Leidos, Inc. It may not be used, reproduced, disclosed, or exported without the written approval of Leidos.
CONTROLLED DATA