feat: allow azure openai to pass custom uri

[umuthopeyildirim]

Title:
feat: allow azure openai to pass custom uri

Description: (optional)
This pull request introduces support for a custom Azure URI for OpenAI requests. The following changes have been made:

Header Handling:
Updated constructConfigFromRequestHeaders to extract the azureCustomUri from the incoming request headers. This enables the passing of a custom URI from the client.

Provider Configuration:
Modified AzureOpenAIAPIConfig in src/providers/azure-openai/api.ts to use the azureCustomUri when constructing the base URL for API requests.
Updated the endpoint resolution logic so that if a custom URI is provided, the URL structure is adapted accordingly (e.g., DeepSeek-R1 on Azure uses {resourceName}.services.ai.azure.com/models).

Type Definitions:
Extended the Options interface in src/types/requestBody.ts to include the azureCustomUri field.

Motivation: (optional)
This enhancement is necessary to support scenarios where a non-standard Azure endpoint is required (such as DeepSeek-R1). By allowing users to pass a custom URI, we provide greater flexibility and ensure compatibility with alternative endpoint configurations.

Related Issues: (optional)

• #issue-number

[VisargD]

This enhancement is necessary to support scenarios where a non-standard Azure endpoint is required (such as DeepSeek-R1). By allowing users to pass a custom URI, we provide greater flexibility and ensure compatibility with alternative endpoint configurations.

Hey @umuthopeyildirim, DeepSeek-R1 (or any other models) are already supported as a part of the azure-ai provider. azure-openai is specifically used for Azure OpenAI services. Here is an example to use azure-ai provider:

curl --request POST \
  --url <gateway_url>/v1/chat/completions \
  --header 'Authorization: API_KEY' \
  --header 'content-type: application/json' \
  --header 'x-portkey-azure-api-version: 2024-05-01-preview' \
  --header 'x-portkey-azure-deployment-name: Llama-2-7b-chat-cgnkg' \
  --header 'x-portkey-azure-deployment-type: serverless' \
  --header 'x-portkey-azure-region: eastus' \
  --header 'x-portkey-provider: azure-ai' \
  --data '{
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant"
    },
    {
      "role": "user",
      "content": "Hello?"
    }
  ]
}'

[umuthopeyildirim]

Hey @VisargD Yes I'm aware of Azure AI Inference but subdomains still don't match + Azure is now deploying DeepSeek-R1 under Azure AI Services. My endpoint for Deepseek is: .services.ai.azure.com

[narengogi]

Hey @umuthopeyildirim I think you could add the same check in the azure-ai provider itself like @VisargD mentioned

I've noticed documentation for this provider is missing, tagging @b4s36t4 who wrote the azure-ai integration to update the docs, and suggest what changes could be made to support deepseek

[b4s36t4]

Hey, @umuthopeyildirim. Thanks for pointing out the issue and the PR.

Although the solution you've proposed is good but I don't think this is applicable for azure-openai. The specific endpoint is only available and can only be created from Azure AI Foundry.

I will create a new PR pointing with your changes, but it's gonna be provider for auzre-ai.

The problem you've faced to integrated Deepseek with azure is not a real problem, the endpoint azure giving is of provider auzre-ai.
you can take a look at the implementation here

gateway/src/providers/azure-ai-inference/api.ts

Lines 16 to 18 in f35fddf

	if (azureDeploymentType === 'serverless') {
	return `https://${azureDeploymentName?.toLowerCase()}.${azureRegion}.models.ai.azure.com`;
	}

.

https://learn.microsoft.com/en-us/azure/ai-foundry/model-inference/concepts/endpoints?tabs=rest A bit more about the problem that relates to azure-ai provider.

Please share your thoughts if I seem diverted from the problem or so.