Merge pull request #17260 from abi-jey/main

fix: GA path for azure openai realtime models

litellm/llms/azure/realtime/handler.py

@@ -12,6 +12,7 @@ from ....litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
 from ....litellm_core_utils.realtime_streaming import RealTimeStreaming
 from ....llms.custom_httpx.http_handler import get_shared_realtime_ssl_context
 from ..azure import AzureChatCompletion
+from litellm._logging import verbose_proxy_logger
 
 # BACKEND_WS_URL = "ws://localhost:8080/v1/realtime?model=gpt-4o-realtime-preview-2024-10-01"
 
@@ -51,18 +52,18 @@ class AzureOpenAIRealtime(AzureChatCompletion):
 
         Examples:
             beta/default: "wss://.../openai/realtime?api-version=2024-10-01-preview&deployment=gpt-4o-realtime-preview"
-            GA/v1:        "wss://.../openai/v1/realtime?api-version=2024-10-01-preview&deployment=gpt-4o-realtime-preview"
+            GA/v1:        "wss://.../openai/v1/realtime?model=gpt-realtime-deployment"
         """
         api_base = api_base.replace("https://", "wss://")
 
         # Determine path based on realtime_protocol
         if realtime_protocol in ("GA", "v1"):
-            path = "/openai/v1/realtime"
+            path = "/openai/v1/realtime" 
+            return f"{api_base}{path}?model={model}"
         else:
             # Default to beta path for backwards compatibility
             path = "/openai/realtime"
-
-        return f"{api_base}{path}?api-version={api_version}&deployment={model}"
+            return f"{api_base}{path}?api-version={api_version}&deployment={model}"
 
     async def async_realtime(
         self,
@@ -107,4 +108,5 @@ class AzureOpenAIRealtime(AzureChatCompletion):
         except websockets.exceptions.InvalidStatusCode as e:  # type: ignore
             await websocket.close(code=e.status_code, reason=str(e))
         except Exception:
+            verbose_proxy_logger.exception("Error in AzureOpenAIRealtime.async_realtime")
             pass

tests/test_litellm/llms/azure/realtime/test_azure_realtime_handler.py

@@ -117,6 +117,7 @@ async def test_construct_url_beta_protocol_explicit():
 async def test_construct_url_ga_protocol():
     """
     Test that realtime_protocol='GA' uses /openai/v1/realtime (GA path).
+    GA path uses ?model= instead of ?api-version=&deployment= format.
     """
     from litellm.llms.azure.realtime.handler import AzureOpenAIRealtime
 
@@ -132,8 +133,10 @@ async def test_construct_url_ga_protocol():
     assert "/openai/v1/realtime?" in url
     # Ensure it doesn't have both paths
     assert url.count("/realtime") == 1
-    assert "api-version=2024-10-01-preview" in url
-    assert "deployment=gpt-4o-realtime-preview" in url
+    # GA path uses model= query param, not api-version and deployment
+    assert "model=gpt-4o-realtime-preview" in url
+    assert "api-version" not in url
+    assert "deployment" not in url
 
 
 @pytest.mark.asyncio
@@ -203,8 +206,10 @@ async def test_async_realtime_uses_ga_protocol_end_to_end():
         called_url = mock_ws_connect.call_args[0][0]
         assert "/openai/v1/realtime" in called_url
         assert called_url.startswith("wss://")
-        assert "api-version=2024-10-01-preview" in called_url
-        assert "deployment=gpt-4o-realtime-preview" in called_url
+        # GA path uses model= query param, not api-version and deployment
+        assert "model=gpt-4o-realtime-preview" in called_url
+        assert "api-version" not in called_url
+        assert "deployment" not in called_url
 
 
 @pytest.mark.asyncio