Hugging Face CEO Clément Delangue proposes training an open-source AI building model using Hugging Face datasets
Will Brown suggested integrating the ecosystem into openenv.
Many users enthusiastically endorsed the Hugging Face CEO's proposal to train an open source AI building model, replying with repeated direct affirmations of approval.
Most Activity
@ClementDelangue let’s get the whole HF ecosystem into openenv 👀
best way to democratize AI research capabilities, every open model going forward gets trained to use HF
Should we try to train an open source AI building model? We obviously have interesting datasets with HF, MLintern, transformers, trl…
@ClementDelangue @_akhaliq Yes!
@ClementDelangue the question isn't whether the data is there - it's whether a model trained on OSS contribution patterns can generalize to the messy ambiguity of actually shipping products
Please do it. You're in one of the best positions in the world to make this happen.
Two things matter: it has to ship fast (we desperately need it), and it has to be something other labs can rally behind over time, Mistral, the Chinese labs, everyone.
If multiple labs and developers actually pull in the same direction, the result could be one for the history books.
@ClementDelangue As many open source companies should link up as possible and try training a model over distributed compute, using something like @NousResearch Psyche
@ClementDelangue another nice thing would be to have a weekly / monthly fine tune on latest research
@ClementDelangue I feel like we 100% should! Even if it is a honest mistake by anthropic I think its important for the research community to make more opensource models again, and not just rely on the good will of anthropic or openai.
@ClementDelangue This is kinda what i was hoping to do with https://huggingface.co/GreatNorthCollective A north american open source community that can better address needs here.
@ClementDelangue Absolutely. It is no longer a question after recent events.
Build the fully open sourced JARVIS. Self-evolving autonomous SETI style distributed compute into cognition. Soon the Agents will dominate over the models. I'm getting happier about this by the day honestly. Let the cards fall where they may. Have this system perfect micro step down re-educating distillation. Genius IQ from tiny language models. Now take this system and go from transformers and rebuild into a diffusion tiny language model (dTLM). This was open AGI at the JARVIS. Then, into running on edge devices. Isn’t this what everyone really wants? Less Earth's maniacs that are going to act evil no matter what.
@ClementDelangue yess!!
@ClementDelangue YES!
@ClementDelangue Yes
@ClementDelangue Yes, I think that would be of great value.
@ClementDelangue It would totally makes sense and ml-intern being powered by it would be the highlight of it.
@ClementDelangue Do it, need of an hour.
@ClementDelangue Yes!!!!!
@ClementDelangue yes, how can i help?
@ClementDelangue Yes, please do! Starting from scratch would be ambitious but why not train on one of the great current open weights models?
@ClementDelangue Yes