Use a TVM Compile Run
Compilation produces a model.so artifact and CORE updates function.so_model.
The next serve uses that reference automatically.
compile_run = function.run(
action="compile",
target_architecture="auto",
wait=True,
)
serve_run = function.run(
action="serve",
served_name="my-model",
wait=True,
)
To serve a different compiled model, set the serve Run's model_path, for example
store://my-project/model/tvm-so/model-so:version.