@alexocheema@exolabs wait, NVIDIA DGX Spark is not supported in exo 1.0? I’m confused since it was claimed to be supported back in October. I think a cluster with a few DGX to prefill and a few M3 ultras to generate should probably be one of the top setups no?
Alex using 2 x Spark for prefill + 2 x Ultras for decode (with 4 x tb5 aggregated rdma) is probably the most cost effective killer cluster for almost every model. If exo can support the sparks clustering as well as it does the Mac’s, using a switch like mikrotik crs812 ddq and each Mac with an atto ns-5102 you can have a full 200gbps switch layer that is data center grade for a fraction of the price.
Why not a national ID like any other country? Makes things much easier, streamlines process and overall it’s better for everyone! I live in the US for 20 years (originally from Europe) and I have always been so confused about the pushback in this country. Maybe something for @DOGE to look into?