Hey there! So today we're releasing SupraBrain-50M, a hybrid language model that combines Gated DeltaNet linear recurrence with Sliding-Window Attention and Surprise-Gated update mechanisms to deliver very strong performance. Here are the benchmarks: https://preview.redd.it/tv0bxkbyh4hh1.png?width=615&format=png&auto=webp&s=f296ed9e6c1d94738cb7aa84015117dd3e65d2a5 Despite being trained on MUCH less data (5B vs 20B tokens!!), it's almost as good as Supra-Base-50M! 🔥 Some s...