diff --git a/litellm/llms/zai/chat/transformation.py b/litellm/llms/zai/chat/transformation.py index bf3e1305d03..3a72d233e6b 100644 --- a/litellm/llms/zai/chat/transformation.py +++ b/litellm/llms/zai/chat/transformation.py @@ -2,7 +2,6 @@ from collections.abc import Mapping from typing import Final import litellm -from litellm.exceptions import UnsupportedParamsError from litellm.secret_managers.main import get_secret_str from litellm.types.llms.openai import AllMessageValues, ChatCompletionToolParam @@ -73,27 +72,11 @@ class ZAIChatConfig(OpenAIGPTConfig): drop_params: bool, ) -> dict[str, object]: # mutable-ok: request pipeline adds parameters to the returned body supported_params: Final = self.get_supported_openai_params(model) - response_format: Final = non_default_params.get("response_format") - unsupported_response_format: Final = response_format is not None and not _is_json_mode(response_format) - if unsupported_response_format and not (drop_params or litellm.drop_params): - raise UnsupportedParamsError( - message=f"zai supports only response_format {{'type': 'json_object'}} for model={model}. To drop it, set `litellm.drop_params=True`", - llm_provider="zai", - model=model, - ) return { **optional_params, **{ ("max_tokens" if param == "max_completion_tokens" else param): value for param, value in non_default_params.items() - if param in supported_params and not (param == "response_format" and unsupported_response_format) + if param in supported_params }, } - - -def _is_json_mode(response_format: object) -> bool: - match response_format: - case {"type": "json_object" | "text"}: - return True - case _: - return False diff --git a/tests/unit/llms/zai/test_zai_provider.py b/tests/unit/llms/zai/test_zai_provider.py index 3963ebc8e95..b3c9e8344a9 100644 --- a/tests/unit/llms/zai/test_zai_provider.py +++ b/tests/unit/llms/zai/test_zai_provider.py @@ -141,36 +141,21 @@ def test_zai_max_completion_tokens_wins_over_max_tokens(respx_mock, zai_response assert body["max_tokens"] == 15 -def test_zai_rejects_json_schema_response_format(respx_mock, monkeypatch): - monkeypatch.setenv("ZAI_API_KEY", "test-api-key") - monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) - route = respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions") - - with pytest.raises(litellm.UnsupportedParamsError) as error: - completion( - model="zai/glm-5v-turbo", - messages=[{"role": "user", "content": "Ann is 30"}], - response_format=JSON_SCHEMA_RESPONSE_FORMAT, - ) - - assert error.value.status_code == 400 - assert not route.called - - -def test_zai_drops_json_schema_response_format_under_drop_params(respx_mock, zai_response, monkeypatch): +@pytest.mark.parametrize("pass_through", [{}, {"allowed_openai_params": ["response_format"]}]) +def test_zai_forwards_json_schema_response_format_as_sent(respx_mock, zai_response, monkeypatch, pass_through): monkeypatch.setenv("ZAI_API_KEY", "test-api-key") monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response) completion( - model="zai/glm-5v-turbo", + model="zai/glm-5.3", messages=[{"role": "user", "content": "Ann is 30"}], response_format=JSON_SCHEMA_RESPONSE_FORMAT, - drop_params=True, + **pass_through, ) body = json.loads(respx_mock.calls[0].request.content) - assert "response_format" not in body + assert body["response_format"] == JSON_SCHEMA_RESPONSE_FORMAT def test_zai_glm_5v_turbo_bills_cached_input_at_cache_read_price(local_model_cost_map: None) -> None: