Pass base parameters to acompletion (#343)

skyvern/forge/sdk/api/llm/api_handler_factory.py

@@ -133,7 +133,7 @@ class LLMAPIHandlerFactory:
         return llm_api_handler_with_router_and_fallback
 
     @staticmethod
-    def get_llm_api_handler(llm_key: str) -> LLMAPIHandler:
+    def get_llm_api_handler(llm_key: str, base_parameters: dict[str, Any] | None = None) -> LLMAPIHandler:
         llm_config = LLMConfigRegistry.get_config(llm_key)
 
         if LLMConfigRegistry.is_router_config(llm_key):
@@ -145,9 +145,12 @@ class LLMAPIHandlerFactory:
             screenshots: list[bytes] | None = None,
             parameters: dict[str, Any] | None = None,
         ) -> dict[str, Any]:
+            active_parameters = base_parameters or {}
             if parameters is None:
                 parameters = LLMAPIHandlerFactory.get_api_parameters()
 
+            active_parameters.update(parameters)
+
             if step:
                 await app.ARTIFACT_MANAGER.create_artifact(
                     step=step,
@@ -174,6 +177,7 @@ class LLMAPIHandlerFactory:
                         {
                             "model": llm_config.model_name,
                             "messages": messages,
+                            # we're not using active_parameters here because it may contain sensitive information
                             **parameters,
                         }
                     ).encode("utf-8"),
@@ -185,7 +189,7 @@ class LLMAPIHandlerFactory:
                 response = await litellm.acompletion(
                     model=llm_config.model_name,
                     messages=messages,
-                    **parameters,
+                    **active_parameters,
                 )
             except openai.OpenAIError as e:
                 raise LLMProviderError(llm_key) from e