My name is Ratko, and I am currently exploring the use of microTVM on RISC-V platforms for a project. I have a crucial question regarding the support of ONNX models in microTVM when executed on RISC-V processors. As I delved into the implementation details, I wanted to ensure compatibility with ONNX models, and I needed help finding specific information in the documentation or tutorials. Could you please clarify whether microTVM, when deployed on RISC-V platforms, supports the conversion and execution of models in the ONNX format?
I would greatly appreciate your insights if there are any considerations, limitations, or best practices related to ONNX support on RISC-V with microTVM. Additionally, any updates or upcoming features in this regard would be valuable information for my ongoing project planning.
Secondly, following the tutorial “9. Bring microTVM to your own development environment,” I encountered an obstacle when I realized that RISC-V is not currently supported. The tutorial has been beneficial up to this point, but unfortunately, I’m experiencing an error that prevents me from proceeding. The error message is as follows:
RuntimeError: schedule_injective not registered for 'c -keys=riscv_cpu ’
when I define target platform as: TARGET = tvm.target.Target(“c -keys=riscv_cpu”) and disable CMSIS_NN.
I have been using Google Colab for this example. Any help would be appreciated.
Greetings, Ratko