chore: Update create_chat_openai_model to include a seed parameter

The `create_chat_openai_model` function in the `openai.py` module has been updated to include a new `seed` parameter. This parameter allows for setting a random seed when creating a `ChatOpenAI` instance. This change improves the control and reproducibility of the generated chat responses.

tests/utils/test_openai.py

@@ -12,7 +12,7 @@
 def test_create_chat_openai_model_with_none(mocker: MockerFixture):
     coai_mock = mocker.patch("werewolf.utils.openai.ChatOpenAI")
     create_chat_openai_model()
-    coai_mock.assert_called_once_with(model=DEFAULT_MODEL)
+    coai_mock.assert_called_once_with(model=DEFAULT_MODEL, seed=None)
 
 
 @pytest.mark.parametrize("model_name", ["gpt-3.5-turbo", "gpt-4o-mini"])
@@ -22,7 +22,7 @@ def test_create_chat_openai_model_with_str(
 ):
     coai_mock = mocker.patch("werewolf.utils.openai.ChatOpenAI")
     create_chat_openai_model(model_name)
-    coai_mock.assert_called_once_with(model=model_name)
+    coai_mock.assert_called_once_with(model=model_name, seed=None)
 
 
 def test_create_chat_openai_model_with_instance(mocker: MockerFixture):
@@ -31,6 +31,29 @@ def test_create_chat_openai_model_with_instance(mocker: MockerFixture):
     assert actual is llm
 
 
+@pytest.mark.parametrize(
+    'llm, seed',
+    [
+        ('gpt-3.5-turbo', 123),
+        ('gpt-4o-mini', 456),
+        ('gpt-4o-mini', None),
+    ],
+)
+def test_create_chat_openai_model_return_same_instance_for_same_input(
+    llm: str,
+    seed: int,
+    mocker: MockerFixture,
+):
+    ChatOpenAI_mock = mocker.patch(
+        "werewolf.utils.openai.ChatOpenAI",
+        return_value=mocker.MagicMock(spec=ChatOpenAI),
+    )
+    actual1 = create_chat_openai_model(llm, seed)
+    actual2 = create_chat_openai_model(llm, seed)
+    assert actual1 is actual2
+    ChatOpenAI_mock.assert_called_once_with(model=llm, seed=seed)
+
+
 @pytest.mark.integration
 @pytest.mark.skipif(
     os.getenv("OPENAI_API_KEY") is None,

werewolf/main.py

@@ -23,6 +23,7 @@
 @click.option('-m', '--model', default=DEFAULT_MODEL, help=f'The model name. Default is {DEFAULT_MODEL}.')  # noqa
 @click.option('-p', '--printer', default='click.echo', help=f'The printer name. The valid values is in {KEYS_FOR_PRINTER}. Default is click.echo.')  # noqa
 @click.option('--sub-model', default=DEFAULT_MODEL, help=f'The sub-model name. Default is {DEFAULT_MODEL}.')  # noqa
+@click.option('--seed', default=None, help='The random seed.')  # noqa
 @click.option('--log-level', default='WARNING', help='The log level, DEBUG, INFO, WARNING, ERROR or CRITICAL. Default is WARNING.')  # noqa
 @click.option('--debug', is_flag=True, help='Whether to show debug logs or not.')  # noqa
 def main(
@@ -38,9 +39,14 @@ def main(
     model: str,
     printer: str,
     sub_model: str,
+    seed: int | None,
     log_level: str,
     debug: bool,
 ):
+    if seed is not None:
+        import random
+        random.seed(seed)
+
     load_dotenv(),
     if os.environ.get('OPENAI_API_KEY') is None:
         raise ValueError('You must set OPENAI_API_KEY in your environment variables or .env file.')  # noqa

werewolf/utils/openai.py

@@ -1,22 +1,26 @@
+from functools import lru_cache
 from langchain_openai import ChatOpenAI
 from ..const import DEFAULT_MODEL
 
 
+@lru_cache(maxsize=None)
 def create_chat_openai_model(
     llm: ChatOpenAI | str | None = None,
+    seed: int | None = None,
 ) -> ChatOpenAI:
     """Create a ChatOpenAI instance.
 
     Args:
         llm (ChatOpenAI | str | None, optional): ChatOpenAI instance or model name. Defaults to None.
+        seed (int, optional): Random seed. Defaults to None.
 
     Returns:
         ChatOpenAI: ChatOpenAI instance
 
     Note:
-
+        seed is used only when llm is a str or None.
     """  # noqa
     if isinstance(llm, str):
-        return ChatOpenAI(model=llm)
+        return ChatOpenAI(model=llm, seed=seed)
     else:
-        return llm or ChatOpenAI(model=DEFAULT_MODEL)
+        return llm or ChatOpenAI(model=DEFAULT_MODEL, seed=seed)