[kurisufag]

It's certainly large enough for trillion-param frontier-tier trainings, which will likely result in capable open-weight models, the thing you just wished for.