integration

adalflow/adalflow/components/agent/runner.py

@@ -1147,7 +1147,8 @@ async def impl_astream(
                         else:
                             # yield the final planner response
                             if output.error is not None:
-                                if "400" in output.error or "429" in output.error: # context too long or rate limite, not recoverable
+                                if "400" in output.error or "429" or "404" in output.error: # context too long or rate limite, not recoverable
+                                    # 404 model not exist
                                     # create a final output item with error and stop the loop
                                     final_output_item = FinalOutputItem(
                                         error=output.error,

adalflow/adalflow/components/model_client/__init__.py

@@ -98,6 +98,13 @@
     OptionalPackages.OPENAI,
 )
 
+# Import utils functions directly (not lazy) since they don't have dependencies
+from adalflow.components.model_client.utils import (
+    process_images_for_response_api,
+    format_content_for_response_api,
+    parse_embedding_response,
+)
+
 __all__ = [
     "CohereAPIClient",
     "TransformerReranker",
@@ -116,6 +123,11 @@
     "XAIClient",
     "FireworksClient",
     "SambaNovaClient",
+    "AzureAIClient",
+    # Utils functions
+    "process_images_for_response_api",
+    "format_content_for_response_api",
+    "parse_embedding_response",
 ]
 
 for name in __all__: