TVM RPC for concurrent inference? (e.g., multi-request server)

this is mainly meant for benchmark and not for concurrent serving.

Your noted path.2 is the better choice