diff --git a/litellm/llms/gemini/realtime/transformation.py b/litellm/llms/gemini/realtime/transformation.py index 79985569c5f..bfe31a89606 100644 --- a/litellm/llms/gemini/realtime/transformation.py +++ b/litellm/llms/gemini/realtime/transformation.py @@ -52,6 +52,7 @@ from litellm.types.llms.vertex_ai import ( GeminiResponseModalities, HttpxBlobType, HttpxContentType, + TranslationConfig, ) from litellm.types.realtime import ( ALL_DELTA_TYPES, @@ -98,6 +99,31 @@ def _gemini_live_speech_config(voice: object) -> Mapping[str, object] | None: return VertexGeminiConfig()._map_audio_params({"voice": voice}) +class _TranslationGenerationConfig(TypedDict, total=False): + translationConfig: ReadOnly[TranslationConfig] + + +def _translation_generation_config(value: object) -> _TranslationGenerationConfig: + empty: Final[_TranslationGenerationConfig] = {} + if not isinstance(value, Mapping): + return empty + target: Final = value.get("target_language_code", value.get("targetLanguageCode")) + if not isinstance(target, str) or not target: + verbose_logger.warning( + "Gemini Realtime: translation_config without target_language_code dropped; " + "live-translate models stream silence without one." + ) + return empty + echo: Final = value.get("echo_target_language", value.get("echoTargetLanguage")) + if isinstance(echo, bool): + with_echo: Final[_TranslationGenerationConfig] = { + "translationConfig": TranslationConfig(targetLanguageCode=target, echoTargetLanguage=echo) + } + return with_echo + config: Final[_TranslationGenerationConfig] = {"translationConfig": TranslationConfig(targetLanguageCode=target)} + return config + + class _GeminiLiveSetupEnvelope(TypedDict, total=False): setup: ReadOnly[BidiGenerateContentSetup] @@ -308,6 +334,7 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): "input_audio_transcription", "turn_detection", "voice", + "translation_config", ] def map_openai_params(self, optional_params: dict, non_default_params: dict) -> dict: @@ -356,6 +383,9 @@ class GeminiRealtimeConfig(BaseRealtimeConfig): speech_config = _gemini_live_speech_config(value) if speech_config: optional_params["generationConfig"]["speechConfig"] = speech_config + optional_params["generationConfig"].update( + _translation_generation_config(non_default_params.get("translation_config")) + ) if len(optional_params["generationConfig"]) == 0: optional_params.pop("generationConfig") return optional_params diff --git a/litellm/types/llms/vertex_ai.py b/litellm/types/llms/vertex_ai.py index ce51e46ef15..e6fd373d1da 100644 --- a/litellm/types/llms/vertex_ai.py +++ b/litellm/types/llms/vertex_ai.py @@ -2,6 +2,7 @@ from enum import Enum from typing import Any, Final, Literal, Protocol from typing_extensions import ( + ReadOnly, Required, TypedDict, ) @@ -231,6 +232,11 @@ class SpeechConfig(TypedDict, total=False): languageCode: str +class TranslationConfig(TypedDict, total=False): + targetLanguageCode: ReadOnly[str] + echoTargetLanguage: ReadOnly[bool] + + class GenerationConfig(TypedDict, total=False): temperature: float top_p: float @@ -252,6 +258,7 @@ class GenerationConfig(TypedDict, total=False): thinkingConfig: GeminiThinkingConfig mediaResolution: str speechConfig: SpeechConfig + translationConfig: ReadOnly[TranslationConfig] class VertexToolName(str, Enum): diff --git a/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py b/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py index bcd5f3d8d19..e2083ea24e9 100644 --- a/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py +++ b/tests/test_litellm/llms/gemini/realtime/test_gemini_realtime_transformation.py @@ -1,6 +1,6 @@ import json from collections.abc import Mapping -from typing import cast +from typing import Final, cast from unittest.mock import MagicMock import pytest @@ -2201,3 +2201,49 @@ def test_gemini_realtime_response_done_reports_no_grounding_when_none_ran(): assert input_details.get("web_search_requests") is None assert input_details.get("google_maps_grounding_requests") is None + + +def test_map_openai_params_translation_config_snake_and_camel(): + """``session.translation_config`` reaches ``generationConfig.translationConfig`` + whether the client spells the keys OpenAI-style or Gemini-style.""" + cfg: Final = GeminiRealtimeConfig() + + snake: Final = cfg.map_openai_params( + optional_params={}, + non_default_params={"translation_config": {"target_language_code": "ko", "echo_target_language": True}}, + ) + assert snake["generationConfig"]["translationConfig"] == {"targetLanguageCode": "ko", "echoTargetLanguage": True} + + camel: Final = cfg.map_openai_params( + optional_params={}, + non_default_params={"translation_config": {"targetLanguageCode": "pl"}}, + ) + assert camel["generationConfig"]["translationConfig"] == {"targetLanguageCode": "pl"} + + +def test_map_openai_params_translation_config_requires_target_language(): + """Without a target language the live-translate models stream silence, so an + incomplete config is dropped rather than forwarded.""" + cfg: Final = GeminiRealtimeConfig() + + for bad in ({}, {"echo_target_language": True}, {"target_language_code": ""}, "ko", None): + mapped: Final = cfg.map_openai_params(optional_params={}, non_default_params={"translation_config": bad}) + assert "translationConfig" not in mapped.get("generationConfig", {}) + + +def test_gemini_session_update_forwards_translation_config_in_setup(): + """End to end: the first session.update carries translationConfig into the Gemini setup.""" + config: Final = GeminiRealtimeConfig() + session_update: Final = { + "type": "session.update", + "session": {"modalities": ["audio"], "translation_config": {"target_language_code": "ko"}}, + } + + messages: Final = config.transform_realtime_request( + json.dumps(session_update), "gemini-3.5-live-translate-preview", session_configuration_request=None + ) + + setup: Final = json.loads(messages[0])["setup"] + assert setup["model"] == "models/gemini-3.5-live-translate-preview" + assert setup["generationConfig"]["translationConfig"] == {"targetLanguageCode": "ko"} + assert setup["generationConfig"]["responseModalities"] == ["AUDIO"]