In the Relay API , some functions seem to suggest we can compile all graphs from ONNX/ tensorflow/ keras etc. frontends, if we can convet them into relay graphs. Can it be validated ?
https://docs.tvm.ai/api/python/relay/frontend.html#tvm.relay.frontend.from_coreml
What support does TVM Relay API provide for, say , ONNX and tensorflow graphs that are constantly changing opeators, or adding some complex control flows in graphs .