How can I integrate TVM into incubator-mxnet?

[levanhung79]

Hi folks,

I am investigating how to integrate TVM into incubator-mxnet (without using/via NNVM) to run incubator-mxnet on a specific board,

It seems possible, following what I read from the below sources:

• How to integrate TVM into Your Project: https://docs.tvm.ai/deploy/integrate.html
• This “bridge” in incubator-mxnet source code: https://github.com/apache/incubator-mxnet/blob/master/src/nnvm/tvm_bridge.cc

So I would like to know: integrate TVM into incubator-mxnet (without using/via NNVM), is it possible?
If so, how can I do it?

Thank you very much!

[ziheng]

For MXNet’s imperative paradigm, here is an example to use tvm_bridge: https://github.com/apache/incubator-mxnet/blob/master/tests/python/gpu/test_tvm_bridge.py#L55

For symbolic paradigm, deploying the MXNet model via NNVM is a good choice for now.