I’m using tvm to convert a tensorflow model. when I call relay.build, there comes a warning,
Cannot find config for target=cuda, workload=(‘conv2d_nchw.cuda’, (‘TENSOR’, (1, 384, 35, 35), ‘float32’), (‘TENSOR’, (224, 384, 1, 1), ’
but in fact, I searched the tf model and can’t find a conv2d node with this input and kernel shape .
when I use auto-tvm, I see the tasks extracted from the model, and I can’t find this workload.
why is this ?
TVM can generate a new conv2d using the conv2ds in a model ?