Implement TextGrad (bounty)

[abrichr]

Feature request

We would like to implement https://github.com/zou-group/textgrad.

Motivation

It works out-of-the-box for a variety of tasks,
where the users only provide the objective function without tuning components or prompts of
the framework.

Bounty

A paid bounty is available. Please suggest a price range 🙏

[kernel-loophole]

@abrichr wanted to work on this in $750.looking forward to discuss more about it .

[abrichr]

Sounds good @kernel-loophole ! Please DM me on Discord 🙏

[kernel-loophole]

@abrichr can you share your discord username

[kernel-loophole]

@abrichr send you request on discord

[kernel-loophole]

@abrichr I will implement TextGrad into OpenAdapt, leveraging its framework for textual gradients to optimize and automate GUI workflows with Large Multimodal Models (LMMs). By integrating TextGrad's natural-language loss function and optimization through textual feedback, I aim to enhance process automation in OpenAdapt.
This will improve task accuracy, reduce repetitive manual input, and optimize performance across various desktop and web-based automation tasks.

[abrichr]

Hi @kernel-loophole , thank you for your interest. Can you please be more precise? e.g. where will you implement textgrad? How will you evaluate whether the implementation has been successful?

[kernel-loophole]

@abrichr currently i do know the repo code architecture but i think where you pass the LLM the input ,the textgrad will be implement in that code where model can learn from backpropagation .and for evaluation we can use the same llm with textgrad and without textgrad ,then compares the performance

[omarsoufiane]

hello @abrichr , you can create a bounty on Bountyhub, it's integrated into github, (a bot will comment to let other people know that there is a bounty) and if no one solves your bounty you can retract it and get your money back, it's a simple process where you import issues by pasting the issue URL and the payment is upfront so that developers attempting to solve the issue know that the amount is already available.