Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Way to sponsor the project #551

Closed
viktor-ferenczi opened this issue Jan 17, 2024 · 5 comments
Closed

Way to sponsor the project #551

viktor-ferenczi opened this issue Jan 17, 2024 · 5 comments

Comments

@viktor-ferenczi
Copy link

What behavior of the library made you think about the improvement?

No sponsorship link provided.

How would you like it to behave?

Please add a way to sponsor the project. A PayPal link would do, for example.

My work depends on getting the constrained generation working well with vLLM. This includes tensor parallel mode, JSON, regex and Lark grammar. Today I got it working (except of the Lark grammar) by cherry-picking changes from PRs and making a fix myself. So it is coming together nicely and I would like to keep the project going.

Thanks.

@SupreethRao99
Copy link

@viktor-ferenczi could you tell us more about how you got contained generation working with tensor parallel mode? a lot of my work also involves a lot of constrained generation.

Thanks!

@viktor-ferenczi
Copy link
Author

@rlouf Could you please review this? Thanks!

@viktor-ferenczi
Copy link
Author

viktor-ferenczi commented Jan 18, 2024

@SupreethRao99

Take a look into this ticket: #524

Found a solution in lapp0's PR: #539

Fixed the problem with that PR, so it works now. Waiting on the PR's author (lapp0) to take the fix and get his PR reviewed.

Until then I keep a working outlines branch here: https://github.com/viktor-ferenczi/outlines/tree/dev

Please note, that my branch includes code from my open PR to change the vLLM request parameters, but you can cherry-pick only the fix to the RegexLogitsProcessor, that's a separate commit.

I use it with 2x4090 GPUs and the TheBloke/deepseek-coder-33B-instruct-AWQ model as I described it in #524. Constrained generation is a huge efficiency boost in many cases, at least for me.

@rlouf
Copy link
Member

rlouf commented Jan 18, 2024

@viktor-ferenczi Thank you so much, it means a lot to us that you're appreciating our work! No need for sponsorship, but we're thinking about ways to get the community more involved in our journey.

@simon-mo
Copy link

From vLLM perspective, we are big fan of Outlines! From our side, we will make sure this is also a priority, please don't hesitate to reach out and let us know what's the blocker.

@rlouf rlouf closed this as not planned Won't fix, can't repro, duplicate, stale Jan 20, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

4 participants