Build and run agents you can see, understand and trust.
Python agentchatbotlarge-language-modelsllmllm-agentmcpmulti-agentmulti-modalreact-agentrealtime-agentvoice-agentvoice-assistant
fix(model): reuse openai.AsyncClient across calls instead of new per call (#2063)
dongfeng3692cba9d48ce4b6
diff --git a/src/agentscope/embedding/_ollama/_model.py b/src/agentscope/embedding/_ollama/_model.pyindex 9210ac22c..ae04111e7 100644--- a/src/agentscope/embedding/_ollama/_model.py+++ b/src/agentscope/embedding/_ollama/_model.py@@ -78,6 +78,10 @@ class OllamaEmbeddingModel(EmbeddingModelBase[str | TextBlock]):self.host: str | None = getattr(credential, "host", None)self.embedding_cache: EmbeddingCacheBase | None = embedding_cache+ import ollama++ self.client: ollama.AsyncClient = ollama.AsyncClient(host=self.host)+async def _call_api(self,inputs: list[str],@@ -112,12 +116,8 @@ class OllamaEmbeddingModel(EmbeddingModelBase[str | TextBlock]):source="cache",)- import ollama-- client = ollama.AsyncClient(host=self.host)-start_time = datetime.now()- response = await client.embed(**api_kwargs)+ response = await self.client.embed(**api_kwargs)time = (datetime.now() - start_time).total_seconds()if self.embedding_cache:diff --git a/src/agentscope/model/_anthropic/_model.py b/src/agentscope/model/_anthropic/_model.pyindex fd7cc39b2..0b1dff8a9 100644--- a/src/agentscope/model/_anthropic/_model.py+++ b/src/agentscope/model/_anthropic/_model.py@@ -108,6 +108,14 @@ class AnthropicChatModel(ChatModelBase):self.formatter = formatter or AnthropicChatFormatter()self.client_kwargs = client_kwargs or {}+ import anthropic++ self.client: anthropic.AsyncAnthropic = anthropic.AsyncAnthropic(+ api_key=self.credential.api_key.get_secret_value(),+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import anthropic@@ -150,16 +158,6 @@ class AnthropicChatModel(ChatModelBase):enabled."""- import anthropic-- client = anthropic.AsyncAnthropic(- **{- "api_key": self.credential.api_key.get_secret_value(),- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-# Anthropic requires max_tokens; fall back to a safe default when# the user hasn't configured one explicitly.max_tokens = self.parameters.max_tokens or 8192@@ -201,7 +199,7 @@ class AnthropicChatModel(ChatModelBase):start_datetime = datetime.now()- response = await client.messages.create(**kwargs)+ response = await self.client.messages.create(**kwargs)if self.stream:return self._parse_anthropic_stream_completion_response(diff --git a/src/agentscope/model/_dashscope/_model.py b/src/agentscope/model/_dashscope/_model.pyindex a7cf78486..e6f48ad41 100644--- a/src/agentscope/model/_dashscope/_model.py+++ b/src/agentscope/model/_dashscope/_model.py@@ -162,6 +162,14 @@ class DashScopeChatModel(ChatModelBase):self.formatter = formatter or DashScopeChatFormatter()self.client_kwargs = client_kwargs or {}+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import openai@@ -197,16 +205,6 @@ class DashScopeChatModel(ChatModelBase):The keyword arguments for DashScope chat completions API,e.g. ``temperature``, ``max_tokens``, ``top_p``, etc."""- import openai-- client = openai.AsyncClient(- **{- "api_key": self.credential.api_key.get_secret_value(),- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)request_kwargs: dict[str, Any] = {@@ -262,7 +260,7 @@ class DashScopeChatModel(ChatModelBase):request_kwargs["stream_options"] = {"include_usage": True}start_datetime = datetime.now()- response = await client.chat.completions.create(**request_kwargs)+ response = await self.client.chat.completions.create(**request_kwargs)if self.stream:return self._parse_stream_response(start_datetime, response)diff --git a/src/agentscope/model/_deepseek/_model.py b/src/agentscope/model/_deepseek/_model.pyindex b04cd31e2..d62a757fc 100644--- a/src/agentscope/model/_deepseek/_model.py+++ b/src/agentscope/model/_deepseek/_model.py@@ -127,6 +127,14 @@ class DeepSeekChatModel(ChatModelBase):self.formatter = formatter or DeepSeekChatFormatter()self.client_kwargs = client_kwargs or {}+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import openai@@ -166,16 +174,6 @@ class DeepSeekChatModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- import openai-- client = openai.AsyncClient(- **{- "api_key": self.credential.api_key.get_secret_value(),- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)kwargs: dict[str, Any] = {@@ -217,7 +215,7 @@ class DeepSeekChatModel(ChatModelBase):kwargs["stream_options"] = {"include_usage": True}start_datetime = datetime.now()- response = await client.chat.completions.create(**kwargs)+ response = await self.client.chat.completions.create(**kwargs)if self.stream:return self._parse_stream_response(start_datetime, response)diff --git a/src/agentscope/model/_gemini/_model.py b/src/agentscope/model/_gemini/_model.pyindex 2df87188f..35e47e5ea 100644--- a/src/agentscope/model/_gemini/_model.py+++ b/src/agentscope/model/_gemini/_model.py@@ -206,6 +206,13 @@ class GeminiChatModel(ChatModelBase):self.formatter = formatter or GeminiChatFormatter()self.client_kwargs = client_kwargs or {}+ from google import genai++ self.client: genai.Client = genai.Client(+ api_key=self.credential.api_key.get_secret_value(),+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:from google.genai import errors@@ -245,15 +252,6 @@ class GeminiChatModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- from google import genai-- client = genai.Client(- **{- "api_key": self.credential.api_key.get_secret_value(),- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)config: dict[str, Any] = {**config_kwargs}@@ -295,25 +293,21 @@ class GeminiChatModel(ChatModelBase):start_datetime = datetime.now()if self.stream:- response = await client.aio.models.generate_content_stream(+ response = await self.client.aio.models.generate_content_stream(**kwargs,)- # Pass client to the generator so the aiohttp session it owns- # stays alive until the stream is fully consumed.return self._parse_stream_response(start_datetime,response,- client,)- response = await client.aio.models.generate_content(**kwargs)+ response = await self.client.aio.models.generate_content(**kwargs)return self._parse_completion_response(start_datetime, response)async def _parse_stream_response(self,start_datetime: datetime,response: Any,- _client: Any = None,) -> AsyncGenerator[ChatResponse, None]:"""Parse the Gemini streaming response.@@ -323,10 +317,6 @@ class GeminiChatModel(ChatModelBase):response (`Any`):The Gemini async stream object from``client.aio.models.generate_content_stream``.- _client (`Any`, optional):- The ``genai.Client`` that produced the stream. Held here so- its aiohttp session is not garbage-collected before the- stream is fully consumed.Yields:`ChatResponse`:diff --git a/src/agentscope/model/_moonshot/_model.py b/src/agentscope/model/_moonshot/_model.pyindex c4dba08e1..a9bcb3178 100644--- a/src/agentscope/model/_moonshot/_model.py+++ b/src/agentscope/model/_moonshot/_model.py@@ -126,6 +126,14 @@ class MoonshotChatModel(ChatModelBase):self.formatter = formatter or MoonshotChatFormatter()self.client_kwargs = client_kwargs or {}+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import openai@@ -165,16 +173,6 @@ class MoonshotChatModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- import openai-- client = openai.AsyncClient(- **{- "api_key": self.credential.api_key.get_secret_value(),- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)kwargs: dict[str, Any] = {@@ -225,7 +223,7 @@ class MoonshotChatModel(ChatModelBase):kwargs["stream_options"] = {"include_usage": True}start_datetime = datetime.now()- response = await client.chat.completions.create(**kwargs)+ response = await self.client.chat.completions.create(**kwargs)if self.stream:return self._parse_stream_response(start_datetime, response)diff --git a/src/agentscope/model/_ollama/_model.py b/src/agentscope/model/_ollama/_model.pyindex a85e0037f..0bc8ff69e 100644--- a/src/agentscope/model/_ollama/_model.py+++ b/src/agentscope/model/_ollama/_model.py@@ -109,6 +109,13 @@ class OllamaChatModel(ChatModelBase):self.formatter = formatter or OllamaChatFormatter()self.client_kwargs = client_kwargs or {}+ import ollama++ self.client: ollama.AsyncClient = ollama.AsyncClient(+ host=self.credential.host,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import httpx@@ -150,15 +157,6 @@ class OllamaChatModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- import ollama-- client = ollama.AsyncClient(- **{- "host": self.credential.host,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)kwargs: dict[str, Any] = {@@ -185,7 +183,7 @@ class OllamaChatModel(ChatModelBase):kwargs["tools"] = fmt_toolsstart_datetime = datetime.now()- response = await client.chat(**kwargs)+ response = await self.client.chat(**kwargs)if self.stream:return self._parse_stream_response(start_datetime, response)diff --git a/src/agentscope/model/_openai_chat/_model.py b/src/agentscope/model/_openai_chat/_model.pyindex d7885189f..87369dcd9 100644--- a/src/agentscope/model/_openai_chat/_model.py+++ b/src/agentscope/model/_openai_chat/_model.py@@ -165,6 +165,15 @@ class OpenAIChatModel(ChatModelBase):self.client_kwargs = client_kwargs or {}self.extra_body = dict(extra_body) if extra_body is not None else None+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ organization=self.credential.organization,+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import openai@@ -204,17 +213,6 @@ class OpenAIChatModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- import openai-- client = openai.AsyncClient(- **{- "api_key": self.credential.api_key.get_secret_value(),- "organization": self.credential.organization,- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)kwargs: dict[str, Any] = {@@ -269,7 +267,7 @@ class OpenAIChatModel(ChatModelBase):kwargs["stream_options"] = {"include_usage": True}start_datetime = datetime.now()- response = await client.chat.completions.create(**kwargs)+ response = await self.client.chat.completions.create(**kwargs)audio_cfg = kwargs.get("audio")audio_fmt = (diff --git a/src/agentscope/model/_openai_response/_model.py b/src/agentscope/model/_openai_response/_model.pyindex b857c02f6..3981e37d8 100644--- a/src/agentscope/model/_openai_response/_model.py+++ b/src/agentscope/model/_openai_response/_model.py@@ -132,6 +132,15 @@ class OpenAIResponseModel(ChatModelBase):self.formatter = formatter or OpenAIResponseFormatter()self.client_kwargs = client_kwargs or {}+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ organization=self.credential.organization,+ base_url=self.credential.base_url,+ **self.client_kwargs,+ )+@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:import openai@@ -171,17 +180,6 @@ class OpenAIResponseModel(ChatModelBase):generator of ``ChatResponse`` objects when streaming isenabled."""- import openai-- client = openai.AsyncClient(- **{- "api_key": self.credential.api_key.get_secret_value(),- "organization": self.credential.organization,- "base_url": self.credential.base_url,- **self.client_kwargs,- },- )-formatted_messages = await self.formatter.format(messages)api_kwargs: dict[str, Any] = {@@ -224,7 +222,7 @@ class OpenAIResponseModel(ChatModelBase):api_kwargs["tool_choice"] = fmt_tool_choicestart_datetime = datetime.now()- response = await client.responses.create(**api_kwargs)+ response = await self.client.responses.create(**api_kwargs)if self.stream:return self._parse_stream_response(start_datetime, response)diff --git a/src/agentscope/tts/_openai/_model.py b/src/agentscope/tts/_openai/_model.pyindex 125333a74..2b45bf983 100644--- a/src/agentscope/tts/_openai/_model.py+++ b/src/agentscope/tts/_openai/_model.py@@ -121,6 +121,14 @@ class OpenAITTSModel(TTSModelBase):stream=stream,)+ import openai++ self.client: openai.AsyncClient = openai.AsyncClient(+ api_key=self.credential.api_key.get_secret_value(),+ organization=self.credential.organization,+ base_url=self.credential.base_url,+ )+async def synthesize(self,text: str | None = None,@@ -142,14 +150,6 @@ class OpenAITTSModel(TTSModelBase):if not text:return TTSResponse(content=None)- import openai-- client = openai.AsyncClient(- api_key=self.credential.api_key.get_secret_value(),- organization=self.credential.organization,- base_url=self.credential.base_url,- )-media_type = _MEDIA_TYPES.get(self.parameters.response_format,_MEDIA_TYPES[_DEFAULT_RESPONSE_FORMAT],@@ -166,9 +166,13 @@ class OpenAITTSModel(TTSModelBase):request_kwargs["instructions"] = self.parameters.instructionsif self.stream:- return self._stream(client, media_type, **request_kwargs)+ return self._stream(self.client, media_type, **request_kwargs)- return await self._aggregate(client, media_type, **request_kwargs)+ return await self._aggregate(+ self.client,+ media_type,+ **request_kwargs,+ )@staticmethodasync def _aggregate(diff --git a/tests/model_anthropic_test.py b/tests/model_anthropic_test.pyindex b469df2d5..87853afc7 100644--- a/tests/model_anthropic_test.py+++ b/tests/model_anthropic_test.py@@ -10,7 +10,7 @@ import jsonfrom typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -112,14 +112,17 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so messages.create() hits it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("anthropic.AsyncAnthropic")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello!"),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_createresult = await self.model([])@@ -132,11 +135,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "msg-1")- @patch("anthropic.AsyncAnthropic")- async def test_tool_call_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_tool_call_response(self) -> None:"""Non-stream tool call response creates ToolCallBlocks."""mock_create = AsyncMock(return_value=_mock_completion(@@ -149,7 +148,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_createresult = await self.model([])@@ -168,11 +167,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):),)- @patch("anthropic.AsyncAnthropic")- async def test_thinking_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_thinking_response(self) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""mock_create = AsyncMock(return_value=_mock_completion(@@ -180,7 +175,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):text="Answer",),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_createresult = await self.model([])@@ -204,11 +199,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):),)- @patch("anthropic.AsyncAnthropic")- async def test_redacted_thinking_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_redacted_thinking_response(self) -> None:"""Non-stream redacted_thinking block is preserved."""redacted = MagicMock()redacted.type = "redacted_thinking"@@ -233,7 +224,7 @@ class TestAnthropicNonStream(IsolatedAsyncioTestCase):resp.usage.cache_read_input_tokens = 0mock_create = AsyncMock(return_value=resp)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_createresult = await self.model([])@@ -274,9 +265,12 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so messages.create() hits it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("anthropic.AsyncAnthropic")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_text(self) -> None:"""Stream text yields n deltas + 1 final with full content."""msg_usage = MagicMock()msg_usage.input_tokens = 10@@ -319,7 +313,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -361,11 +355,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):)self.assertEqual(responses[-1].id, "msg-1")- @patch("anthropic.AsyncAnthropic")- async def test_stream_thinking_and_text(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_thinking_and_text(self) -> None:"""Stream thinking + text yields deltas then final with signature."""msg_usage = MagicMock()msg_usage.input_tokens = 10@@ -418,7 +408,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -476,11 +466,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):],)- @patch("anthropic.AsyncAnthropic")- async def test_stream_redacted_thinking(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_redacted_thinking(self) -> None:"""Stream redacted_thinking block is emitted atcontent_block_start."""msg_usage = MagicMock()@@ -525,7 +511,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -573,11 +559,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):],)- @patch("anthropic.AsyncAnthropic")- async def test_stream_tool_call(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_tool_call(self) -> None:"""Stream tool call yields partial deltas then full accumulatedinput."""msg_usage = MagicMock()@@ -616,7 +598,7 @@ class TestAnthropicStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.messages.create = mock_create+ self.mock_client.messages.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/model_dashscope_test.py b/tests/model_dashscope_test.pyindex 1b7e3a4a2..03a38f0ac 100644--- a/tests/model_dashscope_test.py+++ b/tests/model_dashscope_test.py@@ -10,7 +10,7 @@ import wavefrom typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -167,14 +167,15 @@ class TestDashScopeNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -187,10 +188,8 @@ class TestDashScopeNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "req-1")- @patch("openai.AsyncClient")async def test_tool_call_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream tool call response creates ToolCallBlocks."""mock_create = AsyncMock(@@ -204,7 +203,7 @@ class TestDashScopeNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -223,10 +222,8 @@ class TestDashScopeNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_thinking_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""mock_create = AsyncMock(@@ -235,7 +232,7 @@ class TestDashScopeNonStream(IsolatedAsyncioTestCase):reasoning="Reasoning step...",),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -265,9 +262,10 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_text(self) -> None:"""Stream text yields n deltas + 1 final with full content."""chunks = [_make_stream_chunk(delta_text="Hello"),@@ -278,7 +276,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -319,10 +317,8 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_thinking_and_text(self,- mock_client_cls: MagicMock,) -> None:"""Stream reasoning + text yields deltas then accumulated final."""chunks = [@@ -335,7 +331,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -391,10 +387,8 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_tool_calls(self,- mock_client_cls: MagicMock,) -> None:"""Stream tool call chunks accumulate into final ToolCallBlock."""chunks = [@@ -414,7 +408,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -458,8 +452,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")- async def test_stream_usage(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_usage(self) -> None:"""Stream usage chunk attaches token counts to final response."""chunks = [_make_stream_chunk(delta_text="X"),@@ -469,7 +462,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -490,10 +483,8 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):self.assertEqual(responses[-1].usage.input_tokens, 50)self.assertEqual(responses[-1].usage.output_tokens, 10)- @patch("openai.AsyncClient")async def test_stream_audio_response(self,- mock_client_cls: MagicMock,) -> None:"""Stream PCM deltas produce per-chunk DataBlocks (first chunkprefixed with a streaming WAV header) sharing a stable id, plus a@@ -519,7 +510,7 @@ class TestDashScopeStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/model_deepseek_test.py b/tests/model_deepseek_test.pyindex bc182c136..155c95c90 100644--- a/tests/model_deepseek_test.py+++ b/tests/model_deepseek_test.py@@ -10,7 +10,7 @@ Tests cover both non-streaming and streaming modes, verifying that:from typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -156,14 +156,15 @@ class TestDeepSeekNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello world!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -182,10 +183,8 @@ class TestDeepSeekNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "deepseek-1")- @patch("openai.AsyncClient")async def test_tool_call_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream tool call response creates ToolCallBlocks."""mock_create = AsyncMock(@@ -204,7 +203,7 @@ class TestDeepSeekNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -229,10 +228,8 @@ class TestDeepSeekNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_thinking_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""mock_create = AsyncMock(@@ -241,7 +238,7 @@ class TestDeepSeekNonStream(IsolatedAsyncioTestCase):reasoning="Let me think step by step...",),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -275,11 +272,11 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")async def test_stream_text_response(self,- mock_client_cls: MagicMock,) -> None:"""Stream text yields n deltas (is_last=False) + 1 final(is_last=True) with full content."""@@ -293,7 +290,7 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -339,10 +336,8 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):)self.assertEqual(responses[-1].id, "deepseek-1")- @patch("openai.AsyncClient")async def test_stream_thinking_and_text(self,- mock_client_cls: MagicMock,) -> None:"""Stream with thinking + text yields deltas then final with both."""chunks = [@@ -356,7 +351,7 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -422,10 +417,8 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_tool_calls(self,- mock_client_cls: MagicMock,) -> None:"""Stream tool calls accumulate across chunks into final response."""chunks = [@@ -447,7 +440,7 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -491,10 +484,8 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_text_then_tool_call(self,- mock_client_cls: MagicMock,) -> None:"""Stream with text followed by tool call accumulates both."""chunks = [@@ -517,7 +508,7 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -580,10 +571,8 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_usage_in_final(self,- mock_client_cls: MagicMock,) -> None:"""Usage information is captured and present in final response."""chunks = [@@ -598,7 +587,7 @@ class TestDeepSeekStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/model_gemini_test.py b/tests/model_gemini_test.pyindex fe5f56e60..a95bd7415 100644--- a/tests/model_gemini_test.py+++ b/tests/model_gemini_test.py@@ -9,7 +9,7 @@ import jsonfrom typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -118,12 +118,15 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so generate_content() hits it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("google.genai.Client")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""parts = [_make_part(text="Hello!")]- mock_client_cls.return_value.aio.models.generate_content = AsyncMock(+ self.mock_client.aio.models.generate_content = AsyncMock(return_value=_mock_completion(parts),)@@ -137,11 +140,7 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):),)- @patch("google.genai.Client")- async def test_tool_call_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_tool_call_response(self) -> None:"""Non-stream tool call response creates ToolCallBlocks."""parts = [_make_part(@@ -152,7 +151,7 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):},),]- mock_client_cls.return_value.aio.models.generate_content = AsyncMock(+ self.mock_client.aio.models.generate_content = AsyncMock(return_value=_mock_completion(parts),)@@ -176,11 +175,7 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):),)- @patch("google.genai.Client")- async def test_tool_call_response_without_id(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_tool_call_response_without_id(self) -> None:"""A function call with no id gets a generated id, not a crash."""parts = [_make_part(@@ -191,7 +186,7 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):},),]- mock_client_cls.return_value.aio.models.generate_content = AsyncMock(+ self.mock_client.aio.models.generate_content = AsyncMock(return_value=_mock_completion(parts),)@@ -217,17 +212,13 @@ class TestGeminiNonStream(IsolatedAsyncioTestCase):self.assertIsInstance(result.content[0].id, str)self.assertTrue(result.content[0].id)- @patch("google.genai.Client")- async def test_thinking_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_thinking_response(self) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""parts = [_make_part(text="Let me think...", thought=True),_make_part(text="Answer"),]- mock_client_cls.return_value.aio.models.generate_content = AsyncMock(+ self.mock_client.aio.models.generate_content = AsyncMock(return_value=_mock_completion(parts),)@@ -263,16 +254,20 @@ class TestGeminiStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so generate_content_stream() hits it instead of the+ # network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("google.genai.Client")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_text(self) -> None:"""Stream text yields n deltas + 1 final with full content."""chunks = [_make_stream_chunk([_make_part(text="Hello")]),_make_stream_chunk([_make_part(text=" world")]),]- mock_client_cls.return_value.aio.models.generate_content_stream = (- AsyncMock(return_value=_MockAsyncStream(chunks))+ self.mock_client.aio.models.generate_content_stream = AsyncMock(+ return_value=_MockAsyncStream(chunks),)gen = await self.model([])@@ -314,18 +309,14 @@ class TestGeminiStream(IsolatedAsyncioTestCase):],)- @patch("google.genai.Client")- async def test_stream_thinking_and_text(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_thinking_and_text(self) -> None:"""Stream thinking + text yields deltas then accumulated final."""chunks = [_make_stream_chunk([_make_part(text="Think", thought=True)]),_make_stream_chunk([_make_part(text="Answer")]),]- mock_client_cls.return_value.aio.models.generate_content_stream = (- AsyncMock(return_value=_MockAsyncStream(chunks))+ self.mock_client.aio.models.generate_content_stream = AsyncMock(+ return_value=_MockAsyncStream(chunks),)gen = await self.model([])@@ -372,11 +363,7 @@ class TestGeminiStream(IsolatedAsyncioTestCase):],)- @patch("google.genai.Client")- async def test_stream_tool_call(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_tool_call(self) -> None:"""Stream tool call yields delta then final with same ToolCallBlock."""chunks = [_make_stream_chunk(@@ -391,8 +378,8 @@ class TestGeminiStream(IsolatedAsyncioTestCase):],),]- mock_client_cls.return_value.aio.models.generate_content_stream = (- AsyncMock(return_value=_MockAsyncStream(chunks))+ self.mock_client.aio.models.generate_content_stream = AsyncMock(+ return_value=_MockAsyncStream(chunks),)gen = await self.model([])@@ -412,11 +399,7 @@ class TestGeminiStream(IsolatedAsyncioTestCase):],)- @patch("google.genai.Client")- async def test_stream_tool_calls_without_id(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_tool_calls_without_id(self) -> None:"""Two id-less function calls in one chunk get distinct ids."""chunks = [_make_stream_chunk(@@ -438,8 +421,8 @@ class TestGeminiStream(IsolatedAsyncioTestCase):],),]- mock_client_cls.return_value.aio.models.generate_content_stream = (- AsyncMock(return_value=_MockAsyncStream(chunks))+ self.mock_client.aio.models.generate_content_stream = AsyncMock(+ return_value=_MockAsyncStream(chunks),)gen = await self.model([])diff --git a/tests/model_moonshot_test.py b/tests/model_moonshot_test.pyindex d62953cc1..71dc098d5 100644--- a/tests/model_moonshot_test.py+++ b/tests/model_moonshot_test.py@@ -7,7 +7,7 @@ Tests cover both non-streaming and streaming modes.from typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -149,14 +149,15 @@ class TestMoonshotNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -169,10 +170,8 @@ class TestMoonshotNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "kimi-1")- @patch("openai.AsyncClient")async def test_tool_call_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream tool call response creates ToolCallBlocks."""mock_create = AsyncMock(@@ -186,7 +185,7 @@ class TestMoonshotNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -205,10 +204,8 @@ class TestMoonshotNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_thinking_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""mock_create = AsyncMock(@@ -217,7 +214,7 @@ class TestMoonshotNonStream(IsolatedAsyncioTestCase):reasoning="Step by step...",),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -262,9 +259,10 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_text(self) -> None:"""Stream text yields n deltas (is_last=False) + 1 final(is_last=True) with full content."""chunks = [@@ -276,7 +274,7 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -311,10 +309,8 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_thinking_then_text(self,- mock_client_cls: MagicMock,) -> None:"""Moonshot yields thinking chunks separately before text."""chunks = [@@ -327,7 +323,7 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -383,10 +379,8 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_tool_calls(self,- mock_client_cls: MagicMock,) -> None:"""Stream tool calls accumulate across chunks into final response."""chunks = [@@ -406,7 +400,7 @@ class TestMoonshotStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/model_ollama_test.py b/tests/model_ollama_test.pyindex 84fe3676c..a357b2d25 100644--- a/tests/model_ollama_test.py+++ b/tests/model_ollama_test.py@@ -9,7 +9,7 @@ import jsonfrom typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -119,11 +119,14 @@ class TestOllamaNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so chat() hits it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("ollama.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_mock_completion(content="Hello!"),)@@ -137,13 +140,9 @@ class TestOllamaNonStream(IsolatedAsyncioTestCase):),)- @patch("ollama.AsyncClient")- async def test_tool_call_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_tool_call_response(self) -> None:"""Parsing a tool-call response creates a ToolCallBlock."""- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_mock_completion(tool_calls=[{"name": "get_weather", "args": {"city": "SH"}},@@ -168,14 +167,10 @@ class TestOllamaNonStream(IsolatedAsyncioTestCase):),)- @patch("ollama.AsyncClient")- async def test_thinking_response(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_thinking_response(self) -> None:"""Non-stream thinking plus text returns ThinkingBlock thenTextBlock."""- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_mock_completion(content="Answer",thinking="Let me think...",@@ -214,15 +209,18 @@ class TestOllamaStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so chat() hits it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("ollama.AsyncClient")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:- """Stream text yields deltas then final with full content."""+ async def test_stream_text(self) -> None:+ """Stream text yields deltas then full content."""chunks = [_make_stream_chunk(content="Hi"),_make_stream_chunk(content=" there"),]- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_MockAsyncStream(chunks),)@@ -259,18 +257,14 @@ class TestOllamaStream(IsolatedAsyncioTestCase):],)- @patch("ollama.AsyncClient")- async def test_stream_thinking_and_text(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_thinking_and_text(self) -> None:"""Stream thinking and text deltas then final with accumulatedcontent."""chunks = [_make_stream_chunk(thinking="Think step"),_make_stream_chunk(content="Result"),]- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_MockAsyncStream(chunks),)@@ -318,11 +312,7 @@ class TestOllamaStream(IsolatedAsyncioTestCase):],)- @patch("ollama.AsyncClient")- async def test_stream_tool_call(- self,- mock_client_cls: MagicMock,- ) -> None:+ async def test_stream_tool_call(self) -> None:"""Stream tool-call chunk yields delta then final with sameToolCallBlock."""chunks = [@@ -332,7 +322,7 @@ class TestOllamaStream(IsolatedAsyncioTestCase):],),]- mock_client_cls.return_value.chat = AsyncMock(+ self.mock_client.chat = AsyncMock(return_value=_MockAsyncStream(chunks),)diff --git a/tests/model_openai_chat_test.py b/tests/model_openai_chat_test.pyindex 142d631ce..d4a48402c 100644--- a/tests/model_openai_chat_test.py+++ b/tests/model_openai_chat_test.py@@ -13,7 +13,7 @@ import wavefrom typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -168,14 +168,17 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ # Client is built eagerly in __init__; inject a mock onto the+ # instance so create()/stream() calls hit it instead of the network.+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello world!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -194,25 +197,21 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "resp-1")- @patch("openai.AsyncClient")async def test_default_thinking_enable_not_forwarded(self,- mock_client_cls: MagicMock,) -> None:"""Default parameters do not add provider-specific extra_body."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello world!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createawait self.model([])self.assertNotIn("extra_body", mock_create.call_args.kwargs)- @patch("openai.AsyncClient")async def test_constructor_extra_body_forwarded(self,- mock_client_cls: MagicMock,) -> None:"""Custom request fields are forwarded to OpenAI-compatible APIs."""model = OpenAIChatModel(@@ -222,10 +221,11 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):context_size=128_000,extra_body={"enable_thinking": False},)+ model.client = self.mock_clientmock_create = AsyncMock(return_value=_mock_completion(text="Hello world!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createawait model([])@@ -234,10 +234,8 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):{"enable_thinking": False},)- @patch("openai.AsyncClient")async def test_generate_kwargs_extra_body_overrides_constructor(self,- mock_client_cls: MagicMock,) -> None:"""Per-call extra_body overrides the constructor default."""model = OpenAIChatModel(@@ -247,10 +245,11 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):context_size=128_000,extra_body={"enable_thinking": False},)+ model.client = self.mock_clientmock_create = AsyncMock(return_value=_mock_completion(text="Hello world!"),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createawait model([], extra_body={"custom_option": "value"})@@ -259,10 +258,8 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):{"custom_option": "value"},)- @patch("openai.AsyncClient")async def test_tool_call_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream tool call response creates ToolCallBlocks."""mock_create = AsyncMock(@@ -276,7 +273,7 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -295,10 +292,8 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_audio_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream audio-only output yields transcript TextBlock + audioDataBlock."""@@ -311,7 +306,7 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):},),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -338,10 +333,8 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_thinking_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream response with reasoning creates ThinkingBlock."""mock_create = AsyncMock(@@ -350,7 +343,7 @@ class TestOpenAIChatNonStream(IsolatedAsyncioTestCase):reasoning="Let me think step by step...",),)- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_createresult = await self.model([])@@ -384,11 +377,11 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")async def test_stream_text_response(self,- mock_client_cls: MagicMock,) -> None:"""Stream text yields n deltas (is_last=False) + 1 final(is_last=True) with full content."""@@ -402,7 +395,7 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -448,10 +441,8 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):)self.assertEqual(responses[-1].id, "resp-1")- @patch("openai.AsyncClient")async def test_stream_thinking_and_text(self,- mock_client_cls: MagicMock,) -> None:"""Stream with thinking + text yields deltas then final with both."""chunks = [@@ -464,7 +455,7 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -520,10 +511,8 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_tool_calls(self,- mock_client_cls: MagicMock,) -> None:"""Stream tool calls accumulate across chunks into final response."""chunks = [@@ -543,7 +532,7 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -587,10 +576,8 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_usage_in_final(self,- mock_client_cls: MagicMock,) -> None:"""Usage information is captured and present in final response."""chunks = [@@ -601,7 +588,7 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -622,10 +609,8 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):self.assertEqual(responses[-1].usage.input_tokens, 100)self.assertEqual(responses[-1].usage.output_tokens, 20)- @patch("openai.AsyncClient")async def test_stream_audio_response(self,- mock_client_cls: MagicMock,) -> None:"""Stream PCM deltas produce per-chunk DataBlocks (first chunkprefixed with a streaming WAV header) sharing a stable id, plus a@@ -663,7 +648,7 @@ class TestOpenAIChatStream(IsolatedAsyncioTestCase):),]mock_create = AsyncMock(return_value=_MockAsyncStream(chunks))- mock_client_cls.return_value.chat.completions.create = mock_create+ self.mock_client.chat.completions.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/model_openai_response_test.py b/tests/model_openai_response_test.pyindex 0a18a6731..3bf0de094 100644--- a/tests/model_openai_response_test.py+++ b/tests/model_openai_response_test.py@@ -8,7 +8,7 @@ OpenAI Responses API uses event-based streaming with response.completed.from typing import Anyimport unittestfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom utils import AnyString@@ -131,14 +131,15 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=False)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_text_response(self, mock_client_cls: MagicMock) -> None:+ async def test_text_response(self) -> None:"""Non-stream text response returns a single ChatResponse."""mock_create = AsyncMock(return_value=_mock_completion(text="Hello!"),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_createresult = await self.model([])@@ -151,10 +152,8 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):)self.assertEqual(result.id, "resp-openai-1")- @patch("openai.AsyncClient")async def test_tool_call_response(self,- mock_client_cls: MagicMock,) -> None:"""Parsing a tool-call response stores call_id as ToolCallBlock.id."""mock_create = AsyncMock(@@ -169,7 +168,7 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):],),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_createresult = await self.model([])@@ -188,10 +187,8 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_reasoning_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream reasoning summary plus text returns both block types."""mock_create = AsyncMock(@@ -201,7 +198,7 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):reasoning_id="rs_abc999",),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_createresult = await self.model([])@@ -225,10 +222,8 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):),)- @patch("openai.AsyncClient")async def test_empty_reasoning_summary_response(self,- mock_client_cls: MagicMock,) -> None:"""Non-stream empty reasoning summary still preserves its item id."""mock_create = AsyncMock(@@ -238,7 +233,7 @@ class TestOpenAIResponseNonStream(IsolatedAsyncioTestCase):reasoning_id="rs_empty",),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_createresult = await self.model([])@@ -301,9 +296,10 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):def setUp(self) -> None:self.model = _make_model(stream=True)+ self.mock_client = MagicMock()+ self.model.client = self.mock_client- @patch("openai.AsyncClient")- async def test_stream_text(self, mock_client_cls: MagicMock) -> None:+ async def test_stream_text(self) -> None:"""Stream text yields deltas then final with full content."""completed_resp = MagicMock()completed_resp.id = "resp-1"@@ -328,7 +324,7 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -369,10 +365,8 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_reasoning_and_text(self,- mock_client_cls: MagicMock,) -> None:"""Stream reasoning and text deltas then final withreasoning_item_id."""@@ -403,7 +397,7 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -465,10 +459,8 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_empty_reasoning_summary_keeps_reasoning_item_id(self,- mock_client_cls: MagicMock,) -> None:"""Stream empty reasoning summary still preserves its item id."""reasoning_item = MagicMock()@@ -497,7 +489,7 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_creategen = await self.model([])responses = [r async for r in gen]@@ -545,10 +537,8 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):],)- @patch("openai.AsyncClient")async def test_stream_function_call(self,- mock_client_cls: MagicMock,) -> None:"""Stream function-call events use call_id as ToolCallBlock.id."""fc_item = MagicMock()@@ -586,7 +576,7 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):mock_create = AsyncMock(return_value=_MockAsyncEventStream(events),)- mock_client_cls.return_value.responses.create = mock_create+ self.mock_client.responses.create = mock_creategen = await self.model([])responses = [r async for r in gen]diff --git a/tests/tts_openai_test.py b/tests/tts_openai_test.pyindex a181fbb09..85fc60c73 100644--- a/tests/tts_openai_test.py+++ b/tests/tts_openai_test.py@@ -11,7 +11,7 @@ Covers:import base64from typing import Anyfrom unittest import IsolatedAsyncioTestCase-from unittest.mock import AsyncMock, MagicMock, patch+from unittest.mock import AsyncMock, MagicMockfrom agentscope.credential import OpenAICredentialfrom agentscope.tts import OpenAITTSModel, TTSResponse@@ -69,9 +69,10 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):"""Non-streaming returns a single TTSResponse with the full audio."""client = _make_mock_client(b"AAAABBBBCCCC", [])model = self._make_model(stream=False)-- with patch("openai.AsyncClient", return_value=client):- result = await model.synthesize("Hello world")+ # Client is built eagerly in __init__; inject the mock onto the+ # instance so synthesize() hits it instead of the network.+ model.client = client+ result = await model.synthesize("Hello world")self.assertIsInstance(result, TTSResponse)self.assertEqual(result.content.source.media_type, _MEDIA_TYPE_MP3)@@ -85,23 +86,23 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):"""``synthesize(None)`` returns an empty response without touchingthe API."""model = self._make_model(stream=False)-- with patch("openai.AsyncClient") as mock_client_cls:- result = await model.synthesize(None)+ mock_client = _make_mock_client(b"", [])+ model.client = mock_client+ result = await model.synthesize(None)self.assertIsNone(result.content)- mock_client_cls.assert_not_called()+ mock_client.audio.speech.create.assert_not_called()async def test_empty_string_short_circuits(self) -> None:"""``synthesize("")`` returns an empty response without touchingthe API."""model = self._make_model(stream=False)-- with patch("openai.AsyncClient") as mock_client_cls:- result = await model.synthesize("")+ mock_client = _make_mock_client(b"", [])+ model.client = mock_client+ result = await model.synthesize("")self.assertIsNone(result.content)- mock_client_cls.assert_not_called()+ mock_client.audio.speech.create.assert_not_called()async def test_wav_media_type(self) -> None:"""The media type follows the ``response_format`` parameter."""@@ -110,9 +111,8 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):stream=False,parameters=OpenAITTSModel.Parameters(response_format="wav"),)-- with patch("openai.AsyncClient", return_value=client):- result = await model.synthesize("Hello world")+ model.client = client+ result = await model.synthesize("Hello world")self.assertEqual(result.content.source.media_type, _MEDIA_TYPE_WAV)@@ -120,10 +120,9 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):"""Each streamed byte chunk yields one TTSResponse."""client = _make_mock_client(b"", [b"AAAA", b"BBBB", b"CCCC"])model = self._make_model(stream=True)-- with patch("openai.AsyncClient", return_value=client):- gen = await model.synthesize("Hello world")- chunks = [c async for c in gen]+ model.client = client+ gen = await model.synthesize("Hello world")+ chunks = [c async for c in gen]payloads = [base64.b64decode(c.content.source.data) for c in chunks]self.assertEqual(payloads, [b"AAAA", b"BBBB", b"CCCC"])@@ -140,10 +139,9 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):"""A lone audio chunk is flagged ``is_last=True``."""client = _make_mock_client(b"", [b"ONLYCHUNK"])model = self._make_model(stream=True)-- with patch("openai.AsyncClient", return_value=client):- gen = await model.synthesize("Hello world")- chunks = [c async for c in gen]+ model.client = client+ gen = await model.synthesize("Hello world")+ chunks = [c async for c in gen]self.assertEqual(len(chunks), 1)self.assertTrue(chunks[0].is_last)@@ -157,10 +155,9 @@ class TestOpenAITTSModel(IsolatedAsyncioTestCase):sentinel so consumers can detect EOS."""client = _make_mock_client(b"", [])model = self._make_model(stream=True)-- with patch("openai.AsyncClient", return_value=client):- gen = await model.synthesize("Hello world")- chunks = [c async for c in gen]+ model.client = client+ gen = await model.synthesize("Hello world")+ chunks = [c async for c in gen]self.assertEqual(len(chunks), 1)self.assertIsNone(chunks[0].content)