I saw you are developing relay frontend to compile tensorflow models, and it will support dynamic input shapes, really appreciate for that. My questions are:
- How to use it to compile dynamic input shapes?
- Does it affect the performance of tuned runtime? Can I still use autotvm?