OpenCollective fund for LibreTranslate

General
1

Hey all :wave:

this has been on my mind for a while and wanted to begin a discussion on it.

A bit more than a year ago I applied for the first cohort of the FLOSS fund initiative (2025 FLOSS/fund disbursements) and LibreTranslate was accepted for 10k USD! While the funds haven’t been disbursed yet due to accounting delays, I think sometimes next year these will likely become available and I’ve been thinking about ways to make good use of them.

To that extent, I was thinking of opening an OpenCollective fund for the project and solicit bids for microgrants to work on various areas of the project from the community. This could be bids for improving the software, adding new features, improving performance, training new language models, work on upstream libraries, etc.

Just wanted to gage interest to see if this is something that people would like to have and if given the opportunity, to get an idea for things that people would be interested to work on.

2 Likes
2

I like the idea. I think microgrants is a good way to get people to assign more importance to working on the project that would otherwise slip one’s mind.

1 Like
3

I would be willing to train a new model (language based on community input?) for $1000 similar to what I did for the Zammad sponsored model.

@pierotofy I think you should keep the majority of the $10k to support your work on LibreTranslate. You’re currently doing most of the day to day maintenance and making that financially sustainable for you is important for the project.

Some other ideas for micro grants:

  • Documentation - this is something open source projects often struggle with. Proprietary software often has better documentation because they’re paying people to write it.
  • More support for Indian languages, this is currently a weakness of Argos Translate but something that people have been asking for. I think currently we only support Hindi and Urdu. I could work on this
  • Develop a web app for people to submit their own translation pairs and rate translations. This way we could crowdsource data
  • Create a bug bounty for someone to find and fix the memory leak in LibreTranslate
  • Train a custom Spacy model for sentence boundary detection
  • Better tooling for managing .argosmodel translation models. I’ve noticed most people aren’t comfortable installing a model with Python and this prevents them from doing any sort of custom language model install besides using --load-only. Maybe I could port Argos Translate GUI to Windows.
  • More performance profiling
1 Like
4

These are great suggestions, especially the documentation, models training and bug bounty ones.

1 Like
5

Update: things are moving forward and funds should get disbursed within the next few weeks, crossing fingers.

6

The grant money has been disbursed! After taxes the net amount was $7,920.

I explored OpenCollective as an option to manage the fund, but most fiscal hosts charge fees (10-15%). I don’t see the benefit. I think I’ll manage it for the time being with a separate account. In the upcoming weeks I’ll study ways to make the fund visible so that others know that funding opportunities are available.

@argosopentech I think it would be fantastic to train a new language model. If you already have a language in mind that has been requested, I will defer to you on which one could be useful. Some major ones that are currently missing:

  • Swahili Issue · GitHub (150 million to 200 million speakers)
  • Punjabi (150 million speakers)
  • Marathi (83 million speakers)

Send me an invoice for the cost and a brief e-mail describing the deliverable and we can get this going. :+1:

7

I’m also re-evaluating the bug-bounty program idea in light of the recent increase of AI slop. With tons of open source projects now removing their existing programs, I’m a bit concerned about having an open doors bug-bounty program. I think it could still be useful to have something for existing vetted contributors though.