hello !
if i have a graph and a parameter that trained from keras, i can make network model by using below function
Network = keras.models.model_from_json( loaded_model_json )
Network.load_weights(“Weight.h5”)
Network.compile(loss=‘categorical_crossentropy’ ,optimizer=‘sgd’ ,metrics=[‘accuracy’])
mod , params = tvm.relay.frontend.from_keras( Network, shape_dict )
but i want to replace some layer with tvm.topi.nn.conv2d or custom tvm primitives.
for example when i got network, the network has a summary like below.
i want to replace block1_conv2 with tvm.topi.nn.conv2d and customize it. is it possible or supported in TVM?