From 8d35a00974303934340ce3dfea5124262a3bf681 Mon Sep 17 00:00:00 2001
From: "Jugal D. Bhatt" <55304795+jugaldb@users.noreply.github.com>
Date: Sat, 19 Jul 2025 05:38:35 +0530
Subject: [PATCH] [LLM Translation] Added model name formats (#12745)

* Added model supports

* invert logic

* Added gpt 35 turbo check

* add test check

* fix ruff check
---
 litellm/llms/azure/chat/gpt_transformation.py | 22 +++++++++----------
 .../test_azure_chat_gpt_transformation.py     | 15 +++++++++++++
 2 files changed, 25 insertions(+), 12 deletions(-)

diff --git a/litellm/llms/azure/chat/gpt_transformation.py b/litellm/llms/azure/chat/gpt_transformation.py
index 97a044cf01..78fa91bbd7 100644
--- a/litellm/llms/azure/chat/gpt_transformation.py
+++ b/litellm/llms/azure/chat/gpt_transformation.py
@@ -12,7 +12,6 @@ from litellm.types.llms.azure import (
     API_VERSION_YEAR_SUPPORTED_RESPONSE_FORMAT,
 )
 from litellm.types.utils import ModelResponse
-from litellm.utils import supports_response_schema
 
 from ....exceptions import UnsupportedParamsError
 from ....types.llms.openai import AllMessageValues
@@ -110,23 +109,22 @@ class AzureOpenAIConfig(BaseConfig):
 
     def _is_response_format_supported_model(self, model: str) -> bool:
         """
-        - all 4o models are supported
-        - check if 'supports_response_format' is True from get_model_info
-        - [TODO] support smart retries for 3.5 models (some supported, some not)
+        Determines if the model supports response_format.
+        - Handles Azure deployment names (e.g., azure/gpt-4.1-suffix)
+        - Normalizes model names (e.g., gpt-4-1 -> gpt-4.1)
+        - Strips deployment-specific suffixes
+        - Passes provider to supports_response_schema
+        - Backwards compatible with previous model name patterns
         """
-        if "4o" in model:
-            return True
-
-        # Normalize model name by replacing dashes between numbers with dots
-        # e.g., gpt-4-1 -> gpt-4.1, gpt-3-5-turbo -> gpt-3.5-turbo
         import re
 
+        # Normalize model name: e.g., gpt-3-5-turbo -> gpt-3.5-turbo
         normalized_model = re.sub(r"(\d)-(\d)", r"\1.\2", model)
 
-        if supports_response_schema(normalized_model):
-            return True
+        if "gpt-3.5" in normalized_model or "gpt-35" in model:
+            return False
 
-        return False
+        return True
 
     def _is_response_format_supported_api_version(
         self, api_version_year: str, api_version_month: str
diff --git a/tests/test_litellm/llms/azure/chat/test_azure_chat_gpt_transformation.py b/tests/test_litellm/llms/azure/chat/test_azure_chat_gpt_transformation.py
index c4bf7a3300..ac873710d4 100644
--- a/tests/test_litellm/llms/azure/chat/test_azure_chat_gpt_transformation.py
+++ b/tests/test_litellm/llms/azure/chat/test_azure_chat_gpt_transformation.py
@@ -11,5 +11,20 @@ from litellm.llms.azure.chat.gpt_transformation import AzureOpenAIConfig
 class TestAzureOpenAIConfig:
     def test_is_response_format_supported_model(self):
         config = AzureOpenAIConfig()
+        # New logic: Azure deployment names with suffixes and prefixes
+        assert config._is_response_format_supported_model("azure/gpt-4.1-suffix")
+        assert config._is_response_format_supported_model("gpt-4.1-suffix")
+        assert config._is_response_format_supported_model("azure/gpt-4-1-suffix")
+        assert config._is_response_format_supported_model("gpt-4-1-suffix")
+        # 4o models (should always be supported)
+        assert config._is_response_format_supported_model("gpt-4o")
+        assert config._is_response_format_supported_model("azure/gpt-4o-custom")
+        # Backwards compatibility: base names
         assert config._is_response_format_supported_model("gpt-4.1")
         assert config._is_response_format_supported_model("gpt-4-1")
+        # Negative test: clearly unsupported model
+        assert not config._is_response_format_supported_model("gpt-3.5-turbo")
+        assert not config._is_response_format_supported_model("gpt-3-5-turbo")
+        assert not config._is_response_format_supported_model("gpt-3-5-turbo-suffix")
+        assert not config._is_response_format_supported_model("gpt-35-turbo-suffix")
+        assert not config._is_response_format_supported_model("gpt-35-turbo")