ayeganov edited a comment on issue #9939: URL: https://github.com/apache/tvm/issues/9939#issuecomment-1013786357
@masahi Thank you for the helpful answers. I am going to go and try compiling the model with the VM. But I think I still managed to to be vague enough with my questions to confuse myself with your answers :). Here is what I want to accomplish and currently think I understand: 1. I need to take an existing ONNX model, that performs well on CPU, but uses too many computation resources for me to deploy it in the wild as is, and convert it to TVM runtime utilizing Metal as hardware accelerator. 2. Take the converted model and load it into my own library in C++ I don't care about compiling the model to TVM in C++, as long as I can run it in C++. In fact, I'd prefer to do it in python, because it is easier to script. I did notice this comment in the referenced implementation you linked: ``` # Compile with Relay VM # --------------------- # Note: Currently only CPU target is supported. For x86 target, it is # highly recommended to build TVM with Intel MKL and Intel OpenMP to get # best performance, due to the existence of large dense operator in # torchvision rcnn models. ``` Is this outdated? I'll give this a shot in a few minutes, but wanted to bring that to your attention in case there is some discrepancy between examples and actual implemented functionality. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
