What are the Maintaners' intentions with Guidance?

[freckletonj]

This was my favorite "constrained generation" library, but I'm afraid it's dead. I would like to be wrong.

It's Oct 5, the last commit was >2mo ago, and maintainers have not been involved in replies, vetting Issues nor PRs, nor adding models, nor fixing bugs.

Would you please give a status report of your intentions, or archive this project?

If not, the community should direct their attention at a fork that isn't dead, although I don't know if those maintainers would be serious about carrying the torch. 2 forks have novel commits, and several github stars:

• https://github.com/Maximilian-Winter/guidance adds llama-cpp-python support
• https://github.com/paolorechia/local-guidance incorporates a web ui

If not, some similar players in this space are actively maintained:

• https://github.com/eth-sri/lmql
• https://github.com/outlines-dev/outlines

[OscarIntellico]

Thank you for the suggestion of other alternatives, i was wondering the same thing about the repository status.

[marcotcr]

Not dead - we have been working on a release with lots of big changes (it's taking longer than expected).
Hoping to merge it into master soon

[freckletonj]

@marcotcr Thanks for the reply! I really did prefer this project over alternatives, but then bugs went uninspected, and PRs unmerged.

Could you tell us about the microsoft -> guidance-ai shift? Who runs the project now, and do they see this as a project they'll want to maintain? I think just giving updates, maybe posting about your future dev plans, etc would go along way to keeping the wind in the community's sails.

[marcotcr]

Yes, sorry, a lot has happened. Me and @slundberg switched employers, and Microsoft moved the project out to guidance-ai/guidance so that we could continue working on it (they were very nice).

The switch took a while, and we let the project lapse in the meantime. Since we picked it back up, me and Scott have been working on this branch, and more recently on this one with what we thought would be a quick change, but proved to result in a larger refactor that obviates many of the current issues. I think we're close to being ready to release (as in... in a couple of weeks), at which point our plan is to write a blog post, revamp README, examples, etc, and go through all issues that are still relevant.

The new version should be way better, but we'll see how the community responds :)

[freckletonj]

❤️ Thanks for the update! ❤️

[nhorton]

@marcotcr - really appreciate hearing the update. We have a Ruby wrapping of Guidance (and some more fun stuff coming) at https://github.com/Unsupervisedcom/guidance and would love to contribute to some of the newer efforts too.

[Nick-tenBroeke]

Fantastic to hear this is still in development. There's nothing I found that quite gives me the control over the results of prompt chaining than guidance. Excited to see the update when it comes out.

Syntax highlighting would be a gamechanger so hoping that's on the roadmap. 🙏

[ibehnam]

Fantastic to hear this is still in development. There's nothing I found that quite gives me the control over the results of prompt chaining than guidance. Excited to see the update when it comes out.

Syntax highlighting would be a gamechanger so hoping that's on the roadmap. 🙏

How about LMQL? I've been thinking about using it instead of guidance but I don't know if anyone has tried both to give an unbiased comparison of them.

[Nick-tenBroeke]

Fantastic to hear this is still in development. There's nothing I found that quite gives me the control over the results of prompt chaining than guidance. Excited to see the update when it comes out.
Syntax highlighting would be a gamechanger so hoping that's on the roadmap. 🙏

How about LMQL? I've been thinking about using it instead of guidance but I don't know if anyone has tried both to give an unbiased comparison of them.

Wasn't aware of this package yet, thanks for pointing it out. From what I can see LMQL is limited to completion endpoints/model which would make it less versatile than guidance. There are some interesting ideas there though so I'll definitely check it out.

[freckletonj]

@Nick-tenBroeke guidance works on fill-in-the-middle models? I wasn't aware. I truly wish this project would revive. Last I tried it it was broken for all my use cases, and relevant issues from other users were ignored :(

[freckletonj]

I see @marcotcr recently joined, so I guess this project may have new life at some point? 302a240

Welcome @marcotcr!

[Harsha-Nori]

Hey @freckletonj, thanks for all your interest and passion about Guidance! As Marco mentioned above, we've been focused on completely revamping guidance in the branch here: https://github.com/guidance-ai/guidance/tree/pythonic2, though these changes are still stabilizing (and therefore not ready to pull into main just yet). Lots of progress being made, just not on the default GH homepage view :)

[caillef]

Hey @freckletonj, thanks for all your interest and passion about Guidance! As Marco mentioned above, we've been focused on completely revamping guidance in the branch here: https://github.com/guidance-ai/guidance/tree/pythonic2, though these changes are still stabilizing (and therefore not ready to pull into main just yet). Lots of progress being made, just not on the default GH homepage view :)

Hey, I would love to use the version on the branch. Even if it is not stabilised, can we still test it? Do you have a small example of the new default python file using OpenAI and not LLama?

[marcotcr]

I see @marcotcr recently joined, so I guess this project may have new life at some point? 302a240

Welcome @marcotcr!

Thank you. I was developing guidance from the very beginning, just hadn't joined the board :)

---

We finally have a new release, sorry it took us so long :)