[Bug]: azure_ai foundry models not working

[taralika]

What happened?

Direct Azure AI foundry endpoint model call:

curl --request POST \
  --url https://my-model.eastus2.models.ai.azure.com/chat/completions \
  --header 'Authorization: Bearer xxx' \
  --data '{
	"model": "phi-4",
	"messages": [
		{
			"role": "user",
			"content": "ping"
		}
	],
	"max_tokens": 10
}'

successful response:

{
	"id": "chatcmpl-691aa22b-766b-4431-82e9-b3acccb249f8",
	"object": "chat.completion",
	"created": 1748498263,
	"model": "phi4",
	"choices": [
		{
			"index": 0,
			"message": {
				"role": "assistant",
				"content": "It looks like you're referencing the \"ping\"",
				"tool_calls": [],
				"reasoning_content": null
			},
			"finish_reason": "length"
		}
	],
	"usage": {
		"prompt_tokens": 8,
		"total_tokens": 18,
		"completion_tokens": 10,
		"prompt_tokens_details": null
	}
}

proxy-config.yaml entry:

- model_name: phi-4
    litellm_params:
      model: azure_ai/my-model
      api_base: https://my-model.eastus2.models.ai.azure.com
      api_key: xxx
    model_info:
      mode: chat
      region: "East US 2"
      base_model: azure_ai/Phi-4
      access_groups: ["default-models"]
      health_check_timeout: 1

call via litellm proxy:

curl --request POST \
  --url https://xxx.com/chat/completions \
  --header 'api-key: xxx' \
  --data '{
	"model": "phi-4",
	"messages": [
		{
			"role": "user",
			"content": "ping"
		}
	],
	"max_tokens": 10
}'

error response: 400 Bad Request

{
	"error": {
		"message": "litellm.BadRequestError: Azure_aiException - {\"error\":{\"code\":\"Invalid input\",\"status\":422,\"message\":\"invalid input error\",\"details\":[{\"type\":\"model_attributes_type\",\"loc\":[\"body\"],\"msg\":\"Input should be a valid dictionary or object to extract fields from\",\"input\":\"{\\\"model\\\": \\\"phi-4-dcai\\\", \\\"messages\\\": [{\\\"role\\\": \\\"user\\\", \\\"content\\\": \\\"ping\\\"}], \\\"stream\\\": false, \\\"max_tokens\\\": 10}\"}]}}. Received Model Group=phi-4\nAvailable Model Group Fallbacks=None",
		"type": null,
		"param": null,
		"code": "400"
	}
}

Expected: successful response via litellm proxy same as when directly calling azure ai foundry endpoint

Relevant log output

Are you a ML Ops Team?

Yes

What LiteLLM version are you on ?

1.71.2.dev1

Twitter / LinkedIn details

No response