fix: map response_format to output_format for TTS

response_format was declared as supported but never mapped to
the ModelsLab API parameter. Now maps mp3/wav/ogg to output_format.

Add tests for response_format mapping.

Note: time.sleep() polling still blocks async event loop - this is
a known limitation of sync polling in LiteLLM's TTS path.

litellm/llms/modelslab/text_to_speech/transformation.py

@@ -100,6 +100,12 @@ class ModelsLabTextToSpeechConfig(BaseTextToSpeechConfig):
             except (ValueError, TypeError):
                 pass
 
+        # Map response_format (mp3, wav, etc.)
+        if "response_format" in optional_params:
+            fmt = optional_params["response_format"]
+            if fmt in ("mp3", "wav", "ogg"):
+                mapped["output_format"] = fmt
+
         # Language from extra params
         if "language" in kwargs:
             lang = kwargs["language"]

tests/test_litellm/llms/modelslab/text_to_speech/test_modelslab_tts_transformation.py

@@ -58,6 +58,27 @@ class TestModelsLabTTSTransformation:
         )
         assert params["speed"] == 1.5
 
+    def test_map_openai_params_response_format_mp3(self):
+        """response_format mp3 maps to output_format mp3."""
+        _, params = self.config.map_openai_params(
+            model="default", optional_params={"response_format": "mp3"}, voice="alloy"
+        )
+        assert params["output_format"] == "mp3"
+
+    def test_map_openai_params_response_format_wav(self):
+        """response_format wav maps to output_format wav."""
+        _, params = self.config.map_openai_params(
+            model="default", optional_params={"response_format": "wav"}, voice="alloy"
+        )
+        assert params["output_format"] == "wav"
+
+    def test_map_openai_params_response_format_ignored_for_invalid(self):
+        """Invalid response_format values are ignored."""
+        _, params = self.config.map_openai_params(
+            model="default", optional_params={"response_format": "flac"}, voice="alloy"
+        )
+        assert "output_format" not in params
+
     def test_get_supported_openai_params(self):
         params = self.config.get_supported_openai_params("default")
         assert "voice" in params