Merge pull request 'feat: opus 4.1 and gpt-5 special config' (#458) from spec_config into development

Reviewed-on: https://git.biggo.com/Funmula/dive-mcp-host/pulls/458

Author: LouisFunmula

dive_mcp_host/host/conf/llm.py

@@ -82,7 +82,7 @@ class LLMConfiguration(BaseModel):
 
     base_url: str | None = Field(default=None, alias="baseURL")
     skip_tls_verify: bool | None = Field(default=None)
-    temperature: float | None = Field(default=0)
+    temperature: float | None = Field(default=None)
     top_p: float | None = Field(default=None)
 
     model_config = pydantic_model_config
@@ -150,6 +150,28 @@ def dump_api_key(self, v: SecretStr | None) -> str | None:
         """Serialize the api_key field to plain text."""
         return v.get_secret_value() if v else None
 
+    @model_validator(mode="after")
+    def temperature_top_p(self) -> Self:
+        """Update default headers for large tokens."""
+        if (
+            "claude-opus-4-1" in self.model
+            and self.configuration
+            and self.configuration.temperature
+            and self.configuration.top_p
+        ):
+            self.configuration.top_p = None
+
+        if (
+            "gpt-5" in self.model
+            and self.configuration
+            and (temperature := self.configuration.temperature)
+        ):
+            if temperature > 0:
+                self.configuration.temperature = 1
+            else:
+                self.configuration.temperature = None
+        return self
+
 
 class LLMBedrockConfig(BaseLLMConfig):
     """Configuration for Bedrock LLM models."""

tests/test_models.py

@@ -230,3 +230,104 @@ def test_ollama_skip_tls_verify() -> None:
     )
 
     assert model._client._client._transport._pool._ssl_context.verify_mode == CERT_NONE  # type: ignore
+
+
+def test_opus_4_1_temperature_top_p() -> None:
+    """Test the temperature and top_p."""
+    raw_config: dict = {
+        "modelProvider": "anthropic",
+        "model": "claude-opus-4-1",
+    }
+
+    # test opus 4.1
+    simple_1 = raw_config.copy()
+    llm_config = LLMConfig.model_validate(simple_1)
+
+    # with temperature and top_p
+    simple_2 = raw_config.copy()
+    simple_2["configuration"] = {"temperature": 0.5, "top_p": 0.5}
+    llm_config = LLMConfig.model_validate(simple_2)
+    # should be subset of llm_config.to_load_model_kwargs()
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "top_p" not in kwargs
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 0.5
+
+    # test only temperature
+    simple_3 = raw_config.copy()
+    simple_3["configuration"] = {"temperature": 0.5}
+    llm_config = LLMConfig.model_validate(simple_3)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "top_p" not in kwargs
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 0.5
+
+    # test only top_p
+    simple_4 = raw_config.copy()
+    simple_4["configuration"] = {"top_p": 0.5}
+    llm_config = LLMConfig.model_validate(simple_4)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "top_p" in kwargs
+    assert "temperature" not in kwargs
+    assert kwargs["top_p"] == 0.5
+
+    # oap provider
+    simple_5 = simple_2.copy()
+    simple_5["modelProvider"] = "oap"
+    llm_config = LLMConfig.model_validate(simple_5)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "top_p" not in kwargs
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 0.5
+
+    # test general llm config
+    simple_6 = simple_2.copy()
+    simple_6["model"] = "gpt-4o"
+    llm_config = LLMConfig.model_validate(simple_6)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "top_p" in kwargs
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 0.5
+    assert kwargs["top_p"] == 0.5
+
+
+def test_gpt_5_temperature() -> None:
+    """Test the GPT-5 temperature."""
+    raw_config: dict = {
+        "modelProvider": "openai",
+        "model": "gpt-5",
+    }
+
+    simple_1 = raw_config.copy()
+    llm_config = LLMConfig.model_validate(simple_1)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "temperature" not in kwargs
+
+    simple_2 = raw_config.copy()
+    simple_2["configuration"] = {"temperature": 0.5}
+    llm_config = LLMConfig.model_validate(simple_2)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 1
+
+    simple_3 = raw_config.copy()
+    simple_3["configuration"] = {"temperature": 0}
+    llm_config = LLMConfig.model_validate(simple_3)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "temperature" not in kwargs
+
+    # test oap provider
+    simple_4 = simple_2.copy()
+    simple_4["modelProvider"] = "oap"
+    llm_config = LLMConfig.model_validate(simple_4)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 1
+
+    # test general llm config
+    simple_5 = simple_2.copy()
+    simple_5["model"] = "gpt-4o"
+    llm_config = LLMConfig.model_validate(simple_5)
+    kwargs = llm_config.to_load_model_kwargs()
+    assert "temperature" in kwargs
+    assert kwargs["temperature"] == 0.5