[bilsbie]

What’s the rough idea of how this is possible? I thought you need the parrelism of a gpu

[raihansaputra]

inference has less pressure of parallelism compared to training