Comment by veunes

"Fake it till you make it" in a nutshell. Half the AI wrappers on the market do the exact same thing: they render pretty activity charts that have absolutely zero correlation with actual VRAM consumption or server-side inference latency