Hacker News new | past | comments | ask | show | jobs | submit
jasonjmcghee | on: Gemma 4 12B: A unified, encoder-free multimodal model
There's a CUDA backend for MLX now. Not sure about the maturity.