how quick is load a cache model from ram ?

[duongnv0499]

hi, this is just a question, I set "Maximum number of checkpoints loaded at the same time" to 10 and enable "Only keep one model on device" on the setting of sd webui. Each switching model takes me about 3s to complete (this is much quicker than the normal unload and reload). is this behavior okay? or we can do it quicker, because I saw the command show
done in 2.6s (send model to cpu: 1.8s, send model to device: 0.8s) i think the model only needs to send to cuda, not CPU. sorry for a stupid question, but Hope you help me answer. Thank you all for the great work