Volcengine - thinking param support + Azure - handle more gpt custom naming patterns (#11914)

* fix(volcengine.py): add thinking param support

Closes https://github.com/BerriAI/litellm/issues/11879

* fix(gpt_transformation.py): handle azure custom names - e.g. `gpt-4-1`

Closes https://github.com/BerriAI/litellm/issues/11834

litellm/llms/azure/chat/gpt_transformation.py

@@ -116,7 +116,14 @@ class AzureOpenAIConfig(BaseConfig):
         """
         if "4o" in model:
             return True
-        elif supports_response_schema(model):
+
+        # 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
+
+        normalized_model = re.sub(r"(\d)-(\d)", r"\1.\2", model)
+
+        if supports_response_schema(normalized_model):
             return True
 
         return False

litellm/llms/volcengine.py

@@ -61,4 +61,5 @@ class VolcEngineConfig(OpenAILikeChatConfig):
             "functions",
             "max_retries",
             "extra_headers",
+            "thinking",
         ]  # works across all models

tests/test_litellm/llms/azure/chat/test_azure_chat_gpt_transformation.py

@@ -0,0 +1,15 @@
+import os
+import sys
+
+sys.path.insert(
+    0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../.."))
+)
+
+from litellm.llms.azure.chat.gpt_transformation import AzureOpenAIConfig
+
+
+class TestAzureOpenAIConfig:
+    def test_is_response_format_supported_model(self):
+        config = AzureOpenAIConfig()
+        assert config._is_response_format_supported_model("gpt-4.1")
+        assert config._is_response_format_supported_model("gpt-4-1")

tests/test_litellm/llms/test_volcengine.py

@@ -0,0 +1,38 @@
+import os
+import sys
+
+from pydantic import BaseModel
+
+from litellm.llms.volcengine import VolcEngineConfig
+from litellm.utils import get_optional_params
+
+
+class TestVolcEngineConfig:
+    def test_get_optional_params(self):
+        config = VolcEngineConfig()
+        supported_params = config.get_supported_openai_params(model="doubao-seed-1.6")
+        assert "thinking" in supported_params
+
+        mapped_params = config.map_openai_params(
+            non_default_params={
+                "thinking": {"type": "disabled"},
+            },
+            optional_params={},
+            model="doubao-seed-1.6",
+            drop_params=False,
+        )
+
+        assert mapped_params == {
+            "thinking": {"type": "disabled"},
+        }
+
+        e2e_mapped_params = get_optional_params(
+            model="doubao-seed-1.6",
+            custom_llm_provider="volcengine",
+            thinking={"type": "enabled"},
+            drop_params=False,
+        )
+
+        assert "thinking" in e2e_mapped_params and e2e_mapped_params["thinking"] == {
+            "type": "enabled",
+        }