Hi, VM makes me a bit confused. In my opinion, TVM’s main process: relay frontend translate the AI model,
then relay build to compile the model for target, runtime to run model in target device.
So what is VM’s main purpose?
The VM is the new runtime for Relay, it builds on the existing runtime
namespace, but is an evolution of the graph runtime and can execute a larger set of models than the existing runtime. It is likely that one day we will no longer need the graph runtime, and can fully use the VM.
1 Like
Could you please link to the RFC? I’m having some trouble finding it
Here it is: https://github.com/apache/incubator-tvm/issues/2810 this was back when we were often starting RFCs directly on the issue tracker.