Batch Size is too Large

2

convolution schedules now support inference one input at time (N=1).
You should try one at a time.