Function tool callback #16637
Function tool callback #16637
Conversation
dosubot
bot
added
the
size:L
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
…ann/llama-index-expert into function_tool_callback
|
Hello @logan-markewich, can you check it out? I've done the asynchronous implementation and now everything is up to standard |
| # Execute sync callback, if available | ||
| callback_output = self._run_sync_callback(tool_output) | ||
| if callback_output: | ||
| final_output_content += f" Callback: {callback_output}" |
There was a problem hiding this comment.
Actually, looking at this again, this seems rather opinionated 😅 What's the intended use case for this?
I kind of expected the callback to just either be logging something, or outright modifying the tool output in place.
Appending a string like this seems kind of strange
There was a problem hiding this comment.
I'll explain the use case we applied here to see if it becomes a bit clearer:
I have a tool that sends information to the external system of my company.
The agent is responsible for setting all parameters for the API call, but I can't blindly trust the information constructed by the agent.
So, I use this callback strategy to request user confirmation.
This confirmation returns to the agent's interaction, allowing the agent to decide the next steps.
Here, we apply this rule with interactions between agents and the front end for confirmation, but if you check the example notebook I attached, you can see this flow abstracted in a certain way that's easier to understand.
Basically, I need the callback return to influence the remainder of the agent's flow. I couldn't think of an easier way to adapt the classes without violating any principles. If the user doesn't want the callback to influence the flow, they just don't return anything in the function.
There was a problem hiding this comment.
It seems like to me a better design might be
callback_output: ToolOutput | None = self._run_sync_callback(tool_output)
Basically the callback either returns a new tool output to override the existing, or returns None and the original tool_output is used
There was a problem hiding this comment.
Then its up to the user how the callback changes the tool output
Description
This is a feature that allows applying some human-in-the-loop concepts in FunctionTool.
Basically, a callback function is added that enables the developer to request user input in the middle of an agent interaction, as well as allowing any programmatic action.
New Package?
Did I fill in the
tool.llamahubsection in thepyproject.tomland provide a detailed README.md for my new integration or package?Version Bump?
Did I bump the version in the
pyproject.tomlfile of the package I am updating? (Except for thellama-index-corepackage)Type of Change
Please delete options that are not relevant.
How Has This Been Tested?
Your pull-request will likely not be merged unless it is covered by some form of impactful unit testing.
Suggested Checklist:
make format; make lintto appease the lint gods