Make it easy to get seperate "prints" for individual runs/ users when using Transformers Agent

[MarcSkovMadsen]

Feature request

I have started exploring the new Transformers Agent. And I would like to build a UI to help me speed up the process.

I might be running multiple runs in parallel or have multiple users using my application. I would like to be able to stream the information from the run as it arrives. I would like to store the information in a database containing all the runs I've done.

Currently all the valuable information about the run is printed I.e. you are using print to inform me like below

==Explanation from the agent==
I will use the following  tool: `image_generator` to generate an image.


==Code generated by the agent==
image = image_generator(prompt="rivers and lakes")


==Result==
<PIL.PngImagePlugin.PngImageFile image mode=RGB size=512x512 at 0x7F8DDC11C4C0>

This is for example done in agenst.py

Using print makes it hard for me to distinguish between multiple runs/ users. Especially if run in parallel.

Please provide a simple to use method to stream each run individually. It could be as simple as adding a print (or write) argument to the Agent.run, HFAgent.run and OpenAI.run method.

Alternatively some run_id argument could be provided and printed as well. Then I can split the stream that comes in by run_id. This is less preferred though that this also adds some complexity.

Motivation

This will make it much, much easier to create interesting AI apps.

Your contribution

I might do it 😄 . But I hope someone with knowledge of the code base would do it.

Additional Context

An async .run_async function would also be much appreciated as my UI is built on top of Tornado. This will help me keep the app responsive.

[MarcSkovMadsen]

The solution for me is probably to inspect the run function and then compose the pieces in a way that works better for my app.

[amyeroberts]

cc @sgugger @LysandreJik

[sgugger]

Would the PR mentioned above fix your problem?