Expose the "store" parameter through ModelSettings #357
Conversation
| @@ -30,7 +30,7 @@ class ModelSettings: | |||
| tool_choice: Literal["auto", "required", "none"] | str | None = None | |||
| """The tool choice to use when calling the model.""" | |||
|
|
|||
| parallel_tool_calls: bool | None = False | |||
| parallel_tool_calls: bool | None = None | |||
There was a problem hiding this comment.
@rm-openai this seemed like a typo, since it meant that if you tried to override a different setting at the Runner.run level it would also set this back to False
There was a problem hiding this comment.
actually never mind since an or was used below it's fine. but we should actually not use or below because "0" is false-y in Python, so setting "temperature=0" at the Runner.run level would fail to override temperature at the agent level
src/agents/model_settings.py
Outdated
|
|
||
| new_values = { | ||
| k: getattr(override, k) if getattr(override, k) is not None else getattr(self, k) | ||
| for k in asdict(self) | ||
| } | ||
| return ModelSettings(**new_values) |
There was a problem hiding this comment.
feels like unnecessarily expensive to do an asdict, which creates a copy. Why not just use a helper function that sets the param iff its not None?
There was a problem hiding this comment.
switched to __dataclass_fields__ which does not create a copy, added a test
Closes #173
This will also set stored completions to True by default, encouraging a best practice.