3 ms·inference speed of the models is probably the bottleneckby daralthus 2y agoinference speed of the models is probably the bottleneck