Can pre-trained model ported to VTA?

I noticed that tvm v0.5 had been released.
Can I compile my own models to fit them into FPGA with this version of VTA ?

In theory?
Yes
In practice?
Its not that straightforward.
Right now the only layers which are offloaded to the VTA FPGA part are conv2ds (with relu activation and some saturation and typecasting for quantization).
Also the conv2ds have to be nongrouping and not dilated (and I think some strides are a problem but not sure here).