Merge pull request #197 from vintasoftware/feat/get_llm-models-without-temperature

Allow omitting `temperature` in `get_llm` for models that do not support it

Author: pamella

django_ai_assistant/helpers/assistants.py

@@ -71,8 +71,12 @@ class AIAssistant(abc.ABC):  # noqa: F821
     Should be a valid model name from OpenAI, because the default `get_llm` method uses OpenAI.\n
     `get_llm` can be overridden to use a different LLM implementation.
     """
-    temperature: float = 1.0
-    """Temperature to use for the assistant LLM model.\nDefaults to `1.0`."""
+    temperature: float | None = 1.0
+    """Temperature to use for the assistant LLM model.\n
+    Defaults to `1.0`.\n
+    When `None`, the temperature parameter is omitted when constructing the BaseChatModel
+    in the `get_llm` method.
+    """
     tool_max_concurrency: int = 1
     """Maximum number of tools to run concurrently / in parallel.\nDefaults to `1` (no concurrency)."""
     has_rag: bool = False
@@ -238,14 +242,16 @@ def get_model(self) -> str:
         """
         return self.model
 
-    def get_temperature(self) -> float:
+    def get_temperature(self) -> float | None:
         """Get the temperature to use for the assistant LLM model.
         By default, this is the `temperature` attribute, which is `1.0` by default.\n
         Used by the `get_llm` method to create the LLM instance.\n
-        Override the `temperature` attribute or this method to use a different temperature.
+        Override the `temperature` attribute or this method to use a different temperature.\n
+        Returning `None` is a valid option, particularly for models that do not support
+        temperature control, allowing the parameter to be omitted in the `get_llm` method.\n
 
         Returns:
-            float: The temperature to use for the assistant LLM model.
+            float | None: The temperature to use for the assistant LLM model.
         """
         return self.temperature
 
@@ -271,11 +277,18 @@ def get_llm(self) -> BaseChatModel:
         model = self.get_model()
         temperature = self.get_temperature()
         model_kwargs = self.get_model_kwargs()
-        return ChatOpenAI(
-            model=model,
-            temperature=temperature,
-            model_kwargs=model_kwargs,
-        )
+
+        if temperature is not None:
+            return ChatOpenAI(
+                model=model,
+                temperature=temperature,
+                model_kwargs=model_kwargs,
+            )
+        else:
+            return ChatOpenAI(
+                model=model,
+                model_kwargs=model_kwargs,
+            )
 
     def get_structured_output_llm(self) -> Runnable:
         """Get the LLM model to use for the structured output.

tests/test_helpers/test_assistants.py

@@ -121,7 +121,10 @@ def test_AIAssistant_invoke():
                 id="4",
             ),
             HumanMessage(
-                content="What about tomorrow?", additional_kwargs={}, response_metadata={}, id="5"
+                content="What about tomorrow?",
+                additional_kwargs={},
+                response_metadata={},
+                id="5",
             ),
             AIMessage(
                 content="",
@@ -273,6 +276,120 @@ def tool_a(self, foo: str) -> str:
     ]
 
 
+@patch("django_ai_assistant.helpers.assistants.ChatOpenAI")
+def test_AIAssistant_get_llm_default_temperature(mock_chat_openai):
+    class DefaultTempAssistant(AIAssistant):
+        id = "default_temp_assistant"  # noqa: A003
+        name = "Default Temp Assistant"
+        instructions = "Instructions"
+        model = "gpt-test"
+
+    assistant = DefaultTempAssistant()
+    assistant.get_llm()
+
+    mock_chat_openai.assert_called_once_with(
+        model="gpt-test",
+        temperature=1.0,
+        model_kwargs={},
+    )
+
+    AIAssistant.clear_cls_registry()
+
+
+@patch("django_ai_assistant.helpers.assistants.ChatOpenAI")
+def test_AIAssistant_get_llm_custom_float_temperature(mock_chat_openai):
+    custom_temperature = 0.5
+
+    class CustomFloatTempAssistant(AIAssistant):
+        id = "custom_float_temp_assistant"  # noqa: A003
+        name = "Custom Float Temp Assistant"
+        instructions = "Instructions"
+        model = "gpt-test"
+        temperature = custom_temperature
+
+    assistant = CustomFloatTempAssistant()
+    assistant.get_llm()
+
+    mock_chat_openai.assert_called_once_with(
+        model="gpt-test",
+        temperature=custom_temperature,
+        model_kwargs={},
+    )
+
+    AIAssistant.clear_cls_registry()
+
+
+@patch("django_ai_assistant.helpers.assistants.ChatOpenAI")
+def test_AIAssistant_get_llm_override_get_temperature_with_float(mock_chat_openai):
+    custom_temperature = 0.5
+
+    class OverrideGetFloatTempAssistant(AIAssistant):
+        id = "override_get_float_temp_assistant"  # noqa: A003
+        name = "Override Get Float Temp Assistant"
+        instructions = "Instructions"
+        model = "gpt-test"
+
+        def get_temperature(self) -> float | None:
+            return custom_temperature
+
+    assistant = OverrideGetFloatTempAssistant()
+    assistant.get_llm()
+
+    mock_chat_openai.assert_called_once_with(
+        model="gpt-test",
+        temperature=custom_temperature,
+        model_kwargs={},
+    )
+
+    AIAssistant.clear_cls_registry()
+
+
+@patch("django_ai_assistant.helpers.assistants.ChatOpenAI")
+def test_AIAssistant_get_llm_custom_none_temperature(mock_chat_openai):
+    class CustomNoneTempAssistant(AIAssistant):
+        id = "custom_none_temp_assistant"  # noqa: A003
+        name = "Custom None Temp Assistant"
+        instructions = "Instructions"
+        model = "gpt-test"
+        temperature = None
+
+    assistant = CustomNoneTempAssistant()
+    assistant.get_llm()
+
+    mock_chat_openai.assert_called_once_with(
+        model="gpt-test",
+        model_kwargs={},
+    )
+    _, call_kwargs = mock_chat_openai.call_args
+    assert "temperature" not in call_kwargs
+
+    AIAssistant.clear_cls_registry()
+
+
+@patch("django_ai_assistant.helpers.assistants.ChatOpenAI")
+def test_AIAssistant_get_llm_override_get_temperature_with_none(mock_chat_openai):
+    class OverrideGetNoneTempAssistant(AIAssistant):
+        id = "override_get_none_temp_assistant"  # noqa: A003
+        name = "Override Get None Temp Assistant"
+        instructions = "Instructions"
+        model = "gpt-test"
+
+        def get_temperature(self) -> float | None:
+            return None
+
+    assistant = OverrideGetNoneTempAssistant()
+    assistant.get_llm()
+
+    mock_chat_openai.assert_called_once_with(
+        model="gpt-test",
+        model_kwargs={},
+    )
+    _, call_kwargs = mock_chat_openai.call_args
+    assert "temperature" not in call_kwargs
+
+    AIAssistant.clear_cls_registry()
+
+
 @pytest.mark.vcr
 def test_AIAssistant_pydantic_structured_output():
     from pydantic import BaseModel